Acropolis

Benchmark agreement and redundancy

Spearman rho and Kendall tau-b correlations across quality boards, a clustered benchmark map, near-duplicate leaderboards and effective dimensionality.

Near-redundant range (Spearman rho at least 0.90)Distinct benchmark pairs
73< 0.5990.5–0.7770.7–0.8770.8–0.932>= 0.9

Which benchmarks measure the same thing

Boards are grouped by how alike their model rankings are. Distance is sqrt(2 (1 - rho)) from Spearman rho, grouped by average linkage and cut at an average rho of about 0.7. Pairs need at least 10 shared models; the others are left blank. Agreeing rankings show that the same models lead, not that the tasks are the same.

Clusters
6
plus 2 on their own
Boards
33
332 pairs compared
3-D stress
0.32
rough fit
Blank pairs
196
not enough shared models
rho -1+1Blank: fewer than 10 shared modelsBrackets mark clusters.

Hover or use arrow keys to read a pair.

LMArena Text (style control)LMArena Text (style control) and LMArena Text: rho 0.98, 392 shared modelsLMArena Text (style control) and LMArena Vision (style control): rho 0.98, 125 shared modelsLMArena Text (style control) and LMArena Vision: rho 0.96, 125 shared modelsLMArena Text (style control) and GPQA diamond: rho 0.94, 160 shared modelsLMArena Text (style control) and METR time horizon: rho 0.95, 20 shared modelsLMArena Text (style control) and FrontierMath-2025-02-28-Private: rho 0.85, 57 shared modelsLMArena Text (style control) and HLE: rho 0.86, 36 shared modelsLMArena Text (style control) and ARC-AGI-2: rho 0.85, 80 shared modelsLMArena Text (style control) and Terminal Bench: rho 0.87, 33 shared modelsLMArena Text (style control) and SWE-Bench verified: rho 0.81, 27 shared modelsLMArena Text (style control) and LMArena Search: rho 0.78, 36 shared modelsLMArena Text (style control) and Bug Hunt Bench: rho 0.56, 41 shared modelsLMArena Text (style control) and LMArena Search (style control): rho 0.67, 36 shared modelsLMArena Text (style control) and Vending-Bench 2: rho 0.76, 61 shared modelsLMArena Text (style control) and LMArena Document (style control): rho 0.79, 39 shared modelsLMArena Text (style control) and LMArena Document: rho 0.72, 39 shared modelsLMArena Text (style control) and Vals Index: rho 0.63, 24 shared modelsLMArena Text (style control) and Terminal-Bench 4: rho 0.68, 24 shared modelsLMArena Text (style control) and Code Migration: rho 0.59, 24 shared modelsLMArena Text (style control) and LMArena Agent: rho 0.58, 46 shared modelsLMArena Text (style control) and LMArena WebDev: rho 0.73, 112 shared modelsLMArena Text (style control) and CursorBench: rho 0.62, 18 shared modelsLMArena Text (style control) and EMB: rho 0.72, 24 shared modelsLMArena Text (style control) and FrontierCode: rho 0.39, 38 shared modelsLMArena Text (style control) and Vibe Code Bench: rho 0.42, 24 shared modelsLMArena Text (style control) and Legal Research: rho 0.65, 24 shared modelsLMArena Text (style control) and Tax Agent: rho 0.62, 24 shared modelsLMArena Text (style control) and Finance Agent: rho 0.56, 24 shared modelsLMArena Text (style control) and Legal Agent Benchmark (HLAB-Vals): rho 0.49, 24 shared modelsLMArena Text (style control) and GDPval: rho 0.60, 10 shared modelsLMArena Text (style control) and LMArena Image Edit: not enough shared modelsLMArena Text (style control) and LMArena Text-to-Image: not enough shared modelsLMArena Text and LMArena Text (style control): rho 0.98, 392 shared modelsLMArena TextLMArena Text and LMArena Vision (style control): rho 0.94, 125 shared modelsLMArena Text and LMArena Vision: rho 0.97, 125 shared modelsLMArena Text and GPQA diamond: rho 0.93, 160 shared modelsLMArena Text and METR time horizon: rho 0.95, 20 shared modelsLMArena Text and FrontierMath-2025-02-28-Private: rho 0.83, 57 shared modelsLMArena Text and HLE: rho 0.85, 36 shared modelsLMArena Text and ARC-AGI-2: rho 0.78, 80 shared modelsLMArena Text and Terminal Bench: rho 0.76, 33 shared modelsLMArena Text and SWE-Bench verified: rho 0.74, 27 shared modelsLMArena Text and LMArena Search: rho 0.73, 36 shared modelsLMArena Text and Bug Hunt Bench: rho 0.47, 41 shared modelsLMArena Text and LMArena Search (style control): rho 0.43, 36 shared modelsLMArena Text and Vending-Bench 2: rho 0.62, 61 shared modelsLMArena Text and LMArena Document (style control): rho 0.61, 39 shared modelsLMArena Text and LMArena Document: rho 0.67, 39 shared modelsLMArena Text and Vals Index: rho 0.54, 24 shared modelsLMArena Text and Terminal-Bench 4: rho 0.56, 24 shared modelsLMArena Text and Code Migration: rho 0.48, 24 shared modelsLMArena Text and LMArena Agent: rho 0.47, 46 shared modelsLMArena Text and LMArena WebDev: rho 0.65, 112 shared modelsLMArena Text and CursorBench: rho 0.65, 18 shared modelsLMArena Text and EMB: rho 0.58, 24 shared modelsLMArena Text and FrontierCode: rho 0.17, 38 shared modelsLMArena Text and Vibe Code Bench: rho 0.29, 24 shared modelsLMArena Text and Legal Research: rho 0.68, 24 shared modelsLMArena Text and Tax Agent: rho 0.68, 24 shared modelsLMArena Text and Finance Agent: rho 0.60, 24 shared modelsLMArena Text and Legal Agent Benchmark (HLAB-Vals): rho 0.50, 24 shared modelsLMArena Text and GDPval: rho 0.28, 10 shared modelsLMArena Text and LMArena Image Edit: not enough shared modelsLMArena Text and LMArena Text-to-Image: not enough shared modelsLMArena Vision (style control) and LMArena Text (style control): rho 0.98, 125 shared modelsLMArena Vision (style control) and LMArena Text: rho 0.94, 125 shared modelsLMArena Vision (style control)LMArena Vision (style control) and LMArena Vision: rho 0.98, 153 shared modelsLMArena Vision (style control) and GPQA diamond: rho 0.88, 87 shared modelsLMArena Vision (style control) and METR time horizon: rho 0.95, 15 shared modelsLMArena Vision (style control) and FrontierMath-2025-02-28-Private: rho 0.90, 39 shared modelsLMArena Vision (style control) and HLE: rho 0.86, 30 shared modelsLMArena Vision (style control) and ARC-AGI-2: rho 0.86, 57 shared modelsLMArena Vision (style control) and Terminal Bench: rho 0.82, 18 shared modelsLMArena Vision (style control) and SWE-Bench verified: rho 0.77, 18 shared modelsLMArena Vision (style control) and LMArena Search: rho 0.82, 30 shared modelsLMArena Vision (style control) and Bug Hunt Bench: rho 0.26, 29 shared modelsLMArena Vision (style control) and LMArena Search (style control): rho 0.65, 30 shared modelsLMArena Vision (style control) and Vending-Bench 2: rho 0.57, 38 shared modelsLMArena Vision (style control) and LMArena Document (style control): rho 0.78, 37 shared modelsLMArena Vision (style control) and LMArena Document: rho 0.73, 37 shared modelsLMArena Vision (style control) and Vals Index: rho 0.64, 12 shared modelsLMArena Vision (style control) and Terminal-Bench 4: rho 0.64, 12 shared modelsLMArena Vision (style control) and Code Migration: rho 0.55, 12 shared modelsLMArena Vision (style control) and LMArena Agent: rho 0.51, 32 shared modelsLMArena Vision (style control) and LMArena WebDev: rho 0.59, 63 shared modelsLMArena Vision (style control) and CursorBench: rho 0.47, 12 shared modelsLMArena Vision (style control) and EMB: rho 0.66, 12 shared modelsLMArena Vision (style control) and FrontierCode: rho 0.33, 30 shared modelsLMArena Vision (style control) and Vibe Code Bench: rho 0.50, 12 shared modelsLMArena Vision (style control) and Legal Research: rho 0.44, 12 shared modelsLMArena Vision (style control) and Tax Agent: rho 0.50, 12 shared modelsLMArena Vision (style control) and Finance Agent: rho 0.69, 12 shared modelsLMArena Vision (style control) and Legal Agent Benchmark (HLAB-Vals): rho 0.02, 12 shared modelsLMArena Vision (style control) and GDPval: not enough shared modelsLMArena Vision (style control) and LMArena Image Edit: not enough shared modelsLMArena Vision (style control) and LMArena Text-to-Image: not enough shared modelsLMArena Vision and LMArena Text (style control): rho 0.96, 125 shared modelsLMArena Vision and LMArena Text: rho 0.97, 125 shared modelsLMArena Vision and LMArena Vision (style control): rho 0.98, 153 shared modelsLMArena VisionLMArena Vision and GPQA diamond: rho 0.88, 87 shared modelsLMArena Vision and METR time horizon: rho 0.95, 15 shared modelsLMArena Vision and FrontierMath-2025-02-28-Private: rho 0.92, 39 shared modelsLMArena Vision and HLE: rho 0.90, 30 shared modelsLMArena Vision and ARC-AGI-2: rho 0.83, 57 shared modelsLMArena Vision and Terminal Bench: rho 0.77, 18 shared modelsLMArena Vision and SWE-Bench verified: rho 0.79, 18 shared modelsLMArena Vision and LMArena Search: rho 0.73, 30 shared modelsLMArena Vision and Bug Hunt Bench: rho 0.11, 29 shared modelsLMArena Vision and LMArena Search (style control): rho 0.48, 30 shared modelsLMArena Vision and Vending-Bench 2: rho 0.38, 38 shared modelsLMArena Vision and LMArena Document (style control): rho 0.69, 37 shared modelsLMArena Vision and LMArena Document: rho 0.76, 37 shared modelsLMArena Vision and Vals Index: rho 0.43, 12 shared modelsLMArena Vision and Terminal-Bench 4: rho 0.64, 12 shared modelsLMArena Vision and Code Migration: rho 0.42, 12 shared modelsLMArena Vision and LMArena Agent: rho 0.42, 32 shared modelsLMArena Vision and LMArena WebDev: rho 0.54, 63 shared modelsLMArena Vision and CursorBench: rho 0.57, 12 shared modelsLMArena Vision and EMB: rho 0.43, 12 shared modelsLMArena Vision and FrontierCode: rho 0.26, 30 shared modelsLMArena Vision and Vibe Code Bench: rho 0.27, 12 shared modelsLMArena Vision and Legal Research: rho 0.46, 12 shared modelsLMArena Vision and Tax Agent: rho 0.48, 12 shared modelsLMArena Vision and Finance Agent: rho 0.71, 12 shared modelsLMArena Vision and Legal Agent Benchmark (HLAB-Vals): rho -0.05, 12 shared modelsLMArena Vision and GDPval: not enough shared modelsLMArena Vision and LMArena Image Edit: not enough shared modelsLMArena Vision and LMArena Text-to-Image: not enough shared modelsGPQA diamond and LMArena Text (style control): rho 0.94, 160 shared modelsGPQA diamond and LMArena Text: rho 0.93, 160 shared modelsGPQA diamond and LMArena Vision (style control): rho 0.88, 87 shared modelsGPQA diamond and LMArena Vision: rho 0.88, 87 shared modelsGPQA diamondGPQA diamond and METR time horizon: rho 0.95, 20 shared modelsGPQA diamond and FrontierMath-2025-02-28-Private: rho 0.94, 67 shared modelsGPQA diamond and HLE: rho 0.95, 36 shared modelsGPQA diamond and ARC-AGI-2: rho 0.93, 78 shared modelsGPQA diamond and Terminal Bench: rho 0.86, 27 shared modelsGPQA diamond and SWE-Bench verified: rho 0.75, 30 shared modelsGPQA diamond and LMArena Search: rho 0.67, 30 shared modelsGPQA diamond and Bug Hunt Bench: rho 0.81, 37 shared modelsGPQA diamond and LMArena Search (style control): rho 0.77, 30 shared modelsGPQA diamond and Vending-Bench 2: rho 0.74, 55 shared modelsGPQA diamond and LMArena Document (style control): rho 0.41, 36 shared modelsGPQA diamond and LMArena Document: rho 0.48, 36 shared modelsGPQA diamond and Vals Index: rho 0.85, 22 shared modelsGPQA diamond and Terminal-Bench 4: rho 0.76, 22 shared modelsGPQA diamond and Code Migration: rho 0.80, 22 shared modelsGPQA diamond and LMArena Agent: rho 0.77, 35 shared modelsGPQA diamond and LMArena WebDev: rho 0.82, 78 shared modelsGPQA diamond and CursorBench: rho 0.69, 16 shared modelsGPQA diamond and EMB: rho 0.90, 22 shared modelsGPQA diamond and FrontierCode: rho 0.68, 40 shared modelsGPQA diamond and Vibe Code Bench: rho 0.72, 22 shared modelsGPQA diamond and Legal Research: rho 0.50, 22 shared modelsGPQA diamond and Tax Agent: rho 0.46, 22 shared modelsGPQA diamond and Finance Agent: rho 0.57, 22 shared modelsGPQA diamond and Legal Agent Benchmark (HLAB-Vals): rho 0.19, 22 shared modelsGPQA diamond and GDPval: rho 0.33, 11 shared modelsGPQA diamond and LMArena Image Edit: not enough shared modelsGPQA diamond and LMArena Text-to-Image: not enough shared modelsMETR time horizon and LMArena Text (style control): rho 0.95, 20 shared modelsMETR time horizon and LMArena Text: rho 0.95, 20 shared modelsMETR time horizon and LMArena Vision (style control): rho 0.95, 15 shared modelsMETR time horizon and LMArena Vision: rho 0.95, 15 shared modelsMETR time horizon and GPQA diamond: rho 0.95, 20 shared modelsMETR time horizonMETR time horizon and FrontierMath-2025-02-28-Private: rho 0.91, 14 shared modelsMETR time horizon and HLE: rho 0.90, 13 shared modelsMETR time horizon and ARC-AGI-2: rho 0.93, 10 shared modelsMETR time horizon and Terminal Bench: rho 0.79, 10 shared modelsMETR time horizon and SWE-Bench verified: rho 0.85, 11 shared modelsMETR time horizon and LMArena Search: rho 0.92, 10 shared modelsMETR time horizon and Bug Hunt Bench: not enough shared modelsMETR time horizon and LMArena Search (style control): rho 0.59, 10 shared modelsMETR time horizon and Vending-Bench 2: not enough shared modelsMETR time horizon and LMArena Document (style control): not enough shared modelsMETR time horizon and LMArena Document: not enough shared modelsMETR time horizon and Vals Index: not enough shared modelsMETR time horizon and Terminal-Bench 4: not enough shared modelsMETR time horizon and Code Migration: not enough shared modelsMETR time horizon and LMArena Agent: not enough shared modelsMETR time horizon and LMArena WebDev: not enough shared modelsMETR time horizon and CursorBench: not enough shared modelsMETR time horizon and EMB: not enough shared modelsMETR time horizon and FrontierCode: not enough shared modelsMETR time horizon and Vibe Code Bench: not enough shared modelsMETR time horizon and Legal Research: not enough shared modelsMETR time horizon and Tax Agent: not enough shared modelsMETR time horizon and Finance Agent: not enough shared modelsMETR time horizon and Legal Agent Benchmark (HLAB-Vals): not enough shared modelsMETR time horizon and GDPval: not enough shared modelsMETR time horizon and LMArena Image Edit: not enough shared modelsMETR time horizon and LMArena Text-to-Image: not enough shared modelsFrontierMath-2025-02-28-Private and LMArena Text (style control): rho 0.85, 57 shared modelsFrontierMath-2025-02-28-Private and LMArena Text: rho 0.83, 57 shared modelsFrontierMath-2025-02-28-Private and LMArena Vision (style control): rho 0.90, 39 shared modelsFrontierMath-2025-02-28-Private and LMArena Vision: rho 0.92, 39 shared modelsFrontierMath-2025-02-28-Private and GPQA diamond: rho 0.94, 67 shared modelsFrontierMath-2025-02-28-Private and METR time horizon: rho 0.91, 14 shared modelsFrontierMath-2025-02-28-PrivateFrontierMath-2025-02-28-Private and HLE: rho 0.93, 26 shared modelsFrontierMath-2025-02-28-Private and ARC-AGI-2: rho 0.89, 39 shared modelsFrontierMath-2025-02-28-Private and Terminal Bench: rho 0.86, 25 shared modelsFrontierMath-2025-02-28-Private and SWE-Bench verified: rho 0.79, 27 shared modelsFrontierMath-2025-02-28-Private and LMArena Search: rho 0.71, 17 shared modelsFrontierMath-2025-02-28-Private and Bug Hunt Bench: not enough shared modelsFrontierMath-2025-02-28-Private and LMArena Search (style control): rho 0.62, 17 shared modelsFrontierMath-2025-02-28-Private and Vending-Bench 2: rho 0.69, 27 shared modelsFrontierMath-2025-02-28-Private and LMArena Document (style control): rho 0.61, 18 shared modelsFrontierMath-2025-02-28-Private and LMArena Document: rho 0.49, 18 shared modelsFrontierMath-2025-02-28-Private and Vals Index: not enough shared modelsFrontierMath-2025-02-28-Private and Terminal-Bench 4: not enough shared modelsFrontierMath-2025-02-28-Private and Code Migration: not enough shared modelsFrontierMath-2025-02-28-Private and LMArena Agent: not enough shared modelsFrontierMath-2025-02-28-Private and LMArena WebDev: rho 0.72, 28 shared modelsFrontierMath-2025-02-28-Private and CursorBench: not enough shared modelsFrontierMath-2025-02-28-Private and EMB: not enough shared modelsFrontierMath-2025-02-28-Private and FrontierCode: not enough shared modelsFrontierMath-2025-02-28-Private and Vibe Code Bench: not enough shared modelsFrontierMath-2025-02-28-Private and Legal Research: not enough shared modelsFrontierMath-2025-02-28-Private and Tax Agent: not enough shared modelsFrontierMath-2025-02-28-Private and Finance Agent: not enough shared modelsFrontierMath-2025-02-28-Private and Legal Agent Benchmark (HLAB-Vals): not enough shared modelsFrontierMath-2025-02-28-Private and GDPval: rho 0.44, 11 shared modelsFrontierMath-2025-02-28-Private and LMArena Image Edit: not enough shared modelsFrontierMath-2025-02-28-Private and LMArena Text-to-Image: not enough shared modelsHLE and LMArena Text (style control): rho 0.86, 36 shared modelsHLE and LMArena Text: rho 0.85, 36 shared modelsHLE and LMArena Vision (style control): rho 0.86, 30 shared modelsHLE and LMArena Vision: rho 0.90, 30 shared modelsHLE and GPQA diamond: rho 0.95, 36 shared modelsHLE and METR time horizon: rho 0.90, 13 shared modelsHLE and FrontierMath-2025-02-28-Private: rho 0.93, 26 shared modelsHLEHLE and ARC-AGI-2: rho 0.93, 31 shared modelsHLE and Terminal Bench: rho 0.91, 13 shared modelsHLE and SWE-Bench verified: rho 0.81, 17 shared modelsHLE and LMArena Search: rho 0.79, 15 shared modelsHLE and Bug Hunt Bench: not enough shared modelsHLE and LMArena Search (style control): rho 0.82, 15 shared modelsHLE and Vending-Bench 2: rho 0.58, 16 shared modelsHLE and LMArena Document (style control): rho 0.59, 14 shared modelsHLE and LMArena Document: rho 0.44, 14 shared modelsHLE and Vals Index: not enough shared modelsHLE and Terminal-Bench 4: not enough shared modelsHLE and Code Migration: not enough shared modelsHLE and LMArena Agent: not enough shared modelsHLE and LMArena WebDev: rho 0.82, 18 shared modelsHLE and CursorBench: not enough shared modelsHLE and EMB: not enough shared modelsHLE and FrontierCode: not enough shared modelsHLE and Vibe Code Bench: not enough shared modelsHLE and Legal Research: not enough shared modelsHLE and Tax Agent: not enough shared modelsHLE and Finance Agent: not enough shared modelsHLE and Legal Agent Benchmark (HLAB-Vals): not enough shared modelsHLE and GDPval: not enough shared modelsHLE and LMArena Image Edit: not enough shared modelsHLE and LMArena Text-to-Image: not enough shared modelsARC-AGI-2 and LMArena Text (style control): rho 0.85, 80 shared modelsARC-AGI-2 and LMArena Text: rho 0.78, 80 shared modelsARC-AGI-2 and LMArena Vision (style control): rho 0.86, 57 shared modelsARC-AGI-2 and LMArena Vision: rho 0.83, 57 shared modelsARC-AGI-2 and GPQA diamond: rho 0.93, 78 shared modelsARC-AGI-2 and METR time horizon: rho 0.93, 10 shared modelsARC-AGI-2 and FrontierMath-2025-02-28-Private: rho 0.89, 39 shared modelsARC-AGI-2 and HLE: rho 0.93, 31 shared modelsARC-AGI-2ARC-AGI-2 and Terminal Bench: rho 0.89, 23 shared modelsARC-AGI-2 and SWE-Bench verified: rho 0.91, 22 shared modelsARC-AGI-2 and LMArena Search: rho 0.77, 28 shared modelsARC-AGI-2 and Bug Hunt Bench: rho 0.77, 29 shared modelsARC-AGI-2 and LMArena Search (style control): rho 0.87, 28 shared modelsARC-AGI-2 and Vending-Bench 2: rho 0.79, 41 shared modelsARC-AGI-2 and LMArena Document (style control): rho 0.79, 29 shared modelsARC-AGI-2 and LMArena Document: rho 0.78, 29 shared modelsARC-AGI-2 and Vals Index: rho 0.86, 20 shared modelsARC-AGI-2 and Terminal-Bench 4: rho 0.85, 20 shared modelsARC-AGI-2 and Code Migration: rho 0.83, 20 shared modelsARC-AGI-2 and LMArena Agent: rho 0.82, 29 shared modelsARC-AGI-2 and LMArena WebDev: rho 0.84, 58 shared modelsARC-AGI-2 and CursorBench: rho 0.69, 15 shared modelsARC-AGI-2 and EMB: rho 0.81, 20 shared modelsARC-AGI-2 and FrontierCode: rho 0.71, 33 shared modelsARC-AGI-2 and Vibe Code Bench: rho 0.70, 20 shared modelsARC-AGI-2 and Legal Research: rho 0.59, 20 shared modelsARC-AGI-2 and Tax Agent: rho 0.56, 20 shared modelsARC-AGI-2 and Finance Agent: rho 0.71, 20 shared modelsARC-AGI-2 and Legal Agent Benchmark (HLAB-Vals): rho 0.29, 20 shared modelsARC-AGI-2 and GDPval: rho 0.79, 10 shared modelsARC-AGI-2 and LMArena Image Edit: not enough shared modelsARC-AGI-2 and LMArena Text-to-Image: not enough shared modelsTerminal Bench and LMArena Text (style control): rho 0.87, 33 shared modelsTerminal Bench and LMArena Text: rho 0.76, 33 shared modelsTerminal Bench and LMArena Vision (style control): rho 0.82, 18 shared modelsTerminal Bench and LMArena Vision: rho 0.77, 18 shared modelsTerminal Bench and GPQA diamond: rho 0.86, 27 shared modelsTerminal Bench and METR time horizon: rho 0.79, 10 shared modelsTerminal Bench and FrontierMath-2025-02-28-Private: rho 0.86, 25 shared modelsTerminal Bench and HLE: rho 0.91, 13 shared modelsTerminal Bench and ARC-AGI-2: rho 0.89, 23 shared modelsTerminal BenchTerminal Bench and SWE-Bench verified: rho 0.80, 17 shared modelsTerminal Bench and LMArena Search: rho 0.73, 15 shared modelsTerminal Bench and Bug Hunt Bench: not enough shared modelsTerminal Bench and LMArena Search (style control): rho 0.78, 15 shared modelsTerminal Bench and Vending-Bench 2: rho 0.79, 24 shared modelsTerminal Bench and LMArena Document (style control): rho 0.69, 15 shared modelsTerminal Bench and LMArena Document: rho 0.55, 15 shared modelsTerminal Bench and Vals Index: not enough shared modelsTerminal Bench and Terminal-Bench 4: not enough shared modelsTerminal Bench and Code Migration: not enough shared modelsTerminal Bench and LMArena Agent: not enough shared modelsTerminal Bench and LMArena WebDev: rho 0.71, 30 shared modelsTerminal Bench and CursorBench: not enough shared modelsTerminal Bench and EMB: not enough shared modelsTerminal Bench and FrontierCode: not enough shared modelsTerminal Bench and Vibe Code Bench: not enough shared modelsTerminal Bench and Legal Research: not enough shared modelsTerminal Bench and Tax Agent: not enough shared modelsTerminal Bench and Finance Agent: not enough shared modelsTerminal Bench and Legal Agent Benchmark (HLAB-Vals): not enough shared modelsTerminal Bench and GDPval: not enough shared modelsTerminal Bench and LMArena Image Edit: not enough shared modelsTerminal Bench and LMArena Text-to-Image: not enough shared modelsSWE-Bench verified and LMArena Text (style control): rho 0.81, 27 shared modelsSWE-Bench verified and LMArena Text: rho 0.74, 27 shared modelsSWE-Bench verified and LMArena Vision (style control): rho 0.77, 18 shared modelsSWE-Bench verified and LMArena Vision: rho 0.79, 18 shared modelsSWE-Bench verified and GPQA diamond: rho 0.75, 30 shared modelsSWE-Bench verified and METR time horizon: rho 0.85, 11 shared modelsSWE-Bench verified and FrontierMath-2025-02-28-Private: rho 0.79, 27 shared modelsSWE-Bench verified and HLE: rho 0.81, 17 shared modelsSWE-Bench verified and ARC-AGI-2: rho 0.91, 22 shared modelsSWE-Bench verified and Terminal Bench: rho 0.80, 17 shared modelsSWE-Bench verifiedSWE-Bench verified and LMArena Search: rho 0.64, 15 shared modelsSWE-Bench verified and Bug Hunt Bench: not enough shared modelsSWE-Bench verified and LMArena Search (style control): rho 0.59, 15 shared modelsSWE-Bench verified and Vending-Bench 2: rho 0.60, 23 shared modelsSWE-Bench verified and LMArena Document (style control): rho 0.77, 14 shared modelsSWE-Bench verified and LMArena Document: rho 0.74, 14 shared modelsSWE-Bench verified and Vals Index: not enough shared modelsSWE-Bench verified and Terminal-Bench 4: not enough shared modelsSWE-Bench verified and Code Migration: not enough shared modelsSWE-Bench verified and LMArena Agent: not enough shared modelsSWE-Bench verified and LMArena WebDev: rho 0.78, 24 shared modelsSWE-Bench verified and CursorBench: not enough shared modelsSWE-Bench verified and EMB: not enough shared modelsSWE-Bench verified and FrontierCode: not enough shared modelsSWE-Bench verified and Vibe Code Bench: not enough shared modelsSWE-Bench verified and Legal Research: not enough shared modelsSWE-Bench verified and Tax Agent: not enough shared modelsSWE-Bench verified and Finance Agent: not enough shared modelsSWE-Bench verified and Legal Agent Benchmark (HLAB-Vals): not enough shared modelsSWE-Bench verified and GDPval: not enough shared modelsSWE-Bench verified and LMArena Image Edit: not enough shared modelsSWE-Bench verified and LMArena Text-to-Image: not enough shared modelsLMArena Search and LMArena Text (style control): rho 0.78, 36 shared modelsLMArena Search and LMArena Text: rho 0.73, 36 shared modelsLMArena Search and LMArena Vision (style control): rho 0.82, 30 shared modelsLMArena Search and LMArena Vision: rho 0.73, 30 shared modelsLMArena Search and GPQA diamond: rho 0.67, 30 shared modelsLMArena Search and METR time horizon: rho 0.92, 10 shared modelsLMArena Search and FrontierMath-2025-02-28-Private: rho 0.71, 17 shared modelsLMArena Search and HLE: rho 0.79, 15 shared modelsLMArena Search and ARC-AGI-2: rho 0.77, 28 shared modelsLMArena Search and Terminal Bench: rho 0.73, 15 shared modelsLMArena Search and SWE-Bench verified: rho 0.64, 15 shared modelsLMArena SearchLMArena Search and Bug Hunt Bench: rho 0.45, 13 shared modelsLMArena Search and LMArena Search (style control): rho 0.81, 41 shared modelsLMArena Search and Vending-Bench 2: rho 0.61, 26 shared modelsLMArena Search and LMArena Document (style control): rho 0.72, 21 shared modelsLMArena Search and LMArena Document: rho 0.70, 21 shared modelsLMArena Search and Vals Index: not enough shared modelsLMArena Search and Terminal-Bench 4: not enough shared modelsLMArena Search and Code Migration: not enough shared modelsLMArena Search and LMArena Agent: rho 0.08, 12 shared modelsLMArena Search and LMArena WebDev: rho 0.69, 31 shared modelsLMArena Search and CursorBench: not enough shared modelsLMArena Search and EMB: not enough shared modelsLMArena Search and FrontierCode: rho 0.10, 16 shared modelsLMArena Search and Vibe Code Bench: not enough shared modelsLMArena Search and Legal Research: not enough shared modelsLMArena Search and Tax Agent: not enough shared modelsLMArena Search and Finance Agent: not enough shared modelsLMArena Search and Legal Agent Benchmark (HLAB-Vals): not enough shared modelsLMArena Search and GDPval: not enough shared modelsLMArena Search and LMArena Image Edit: not enough shared modelsLMArena Search and LMArena Text-to-Image: not enough shared modelsBug Hunt Bench and LMArena Text (style control): rho 0.56, 41 shared modelsBug Hunt Bench and LMArena Text: rho 0.47, 41 shared modelsBug Hunt Bench and LMArena Vision (style control): rho 0.26, 29 shared modelsBug Hunt Bench and LMArena Vision: rho 0.11, 29 shared modelsBug Hunt Bench and GPQA diamond: rho 0.81, 37 shared modelsBug Hunt Bench and METR time horizon: not enough shared modelsBug Hunt Bench and FrontierMath-2025-02-28-Private: not enough shared modelsBug Hunt Bench and HLE: not enough shared modelsBug Hunt Bench and ARC-AGI-2: rho 0.77, 29 shared modelsBug Hunt Bench and Terminal Bench: not enough shared modelsBug Hunt Bench and SWE-Bench verified: not enough shared modelsBug Hunt Bench and LMArena Search: rho 0.45, 13 shared modelsBug Hunt BenchBug Hunt Bench and LMArena Search (style control): rho 0.86, 13 shared modelsBug Hunt Bench and Vending-Bench 2: rho 0.73, 27 shared modelsBug Hunt Bench and LMArena Document (style control): rho 0.55, 16 shared modelsBug Hunt Bench and LMArena Document: rho 0.45, 16 shared modelsBug Hunt Bench and Vals Index: rho 0.87, 23 shared modelsBug Hunt Bench and Terminal-Bench 4: rho 0.80, 23 shared modelsBug Hunt Bench and Code Migration: rho 0.84, 23 shared modelsBug Hunt Bench and LMArena Agent: rho 0.80, 37 shared modelsBug Hunt Bench and LMArena WebDev: rho 0.81, 44 shared modelsBug Hunt Bench and CursorBench: rho 0.69, 20 shared modelsBug Hunt Bench and EMB: rho 0.69, 23 shared modelsBug Hunt Bench and FrontierCode: rho 0.69, 28 shared modelsBug Hunt Bench and Vibe Code Bench: rho 0.78, 23 shared modelsBug Hunt Bench and Legal Research: rho 0.27, 23 shared modelsBug Hunt Bench and Tax Agent: rho 0.20, 23 shared modelsBug Hunt Bench and Finance Agent: rho 0.39, 23 shared modelsBug Hunt Bench and Legal Agent Benchmark (HLAB-Vals): rho 0.01, 23 shared modelsBug Hunt Bench and GDPval: not enough shared modelsBug Hunt Bench and LMArena Image Edit: not enough shared modelsBug Hunt Bench and LMArena Text-to-Image: not enough shared modelsLMArena Search (style control) and LMArena Text (style control): rho 0.67, 36 shared modelsLMArena Search (style control) and LMArena Text: rho 0.43, 36 shared modelsLMArena Search (style control) and LMArena Vision (style control): rho 0.65, 30 shared modelsLMArena Search (style control) and LMArena Vision: rho 0.48, 30 shared modelsLMArena Search (style control) and GPQA diamond: rho 0.77, 30 shared modelsLMArena Search (style control) and METR time horizon: rho 0.59, 10 shared modelsLMArena Search (style control) and FrontierMath-2025-02-28-Private: rho 0.62, 17 shared modelsLMArena Search (style control) and HLE: rho 0.82, 15 shared modelsLMArena Search (style control) and ARC-AGI-2: rho 0.87, 28 shared modelsLMArena Search (style control) and Terminal Bench: rho 0.78, 15 shared modelsLMArena Search (style control) and SWE-Bench verified: rho 0.59, 15 shared modelsLMArena Search (style control) and LMArena Search: rho 0.81, 41 shared modelsLMArena Search (style control) and Bug Hunt Bench: rho 0.86, 13 shared modelsLMArena Search (style control)LMArena Search (style control) and Vending-Bench 2: rho 0.74, 26 shared modelsLMArena Search (style control) and LMArena Document (style control): rho 0.81, 21 shared modelsLMArena Search (style control) and LMArena Document: rho 0.74, 21 shared modelsLMArena Search (style control) and Vals Index: not enough shared modelsLMArena Search (style control) and Terminal-Bench 4: not enough shared modelsLMArena Search (style control) and Code Migration: not enough shared modelsLMArena Search (style control) and LMArena Agent: rho 0.37, 12 shared modelsLMArena Search (style control) and LMArena WebDev: rho 0.84, 31 shared modelsLMArena Search (style control) and CursorBench: not enough shared modelsLMArena Search (style control) and EMB: not enough shared modelsLMArena Search (style control) and FrontierCode: rho 0.51, 16 shared modelsLMArena Search (style control) and Vibe Code Bench: not enough shared modelsLMArena Search (style control) and Legal Research: not enough shared modelsLMArena Search (style control) and Tax Agent: not enough shared modelsLMArena Search (style control) and Finance Agent: not enough shared modelsLMArena Search (style control) and Legal Agent Benchmark (HLAB-Vals): not enough shared modelsLMArena Search (style control) and GDPval: not enough shared modelsLMArena Search (style control) and LMArena Image Edit: not enough shared modelsLMArena Search (style control) and LMArena Text-to-Image: not enough shared modelsVending-Bench 2 and LMArena Text (style control): rho 0.76, 61 shared modelsVending-Bench 2 and LMArena Text: rho 0.62, 61 shared modelsVending-Bench 2 and LMArena Vision (style control): rho 0.57, 38 shared modelsVending-Bench 2 and LMArena Vision: rho 0.38, 38 shared modelsVending-Bench 2 and GPQA diamond: rho 0.74, 55 shared modelsVending-Bench 2 and METR time horizon: not enough shared modelsVending-Bench 2 and FrontierMath-2025-02-28-Private: rho 0.69, 27 shared modelsVending-Bench 2 and HLE: rho 0.58, 16 shared modelsVending-Bench 2 and ARC-AGI-2: rho 0.79, 41 shared modelsVending-Bench 2 and Terminal Bench: rho 0.79, 24 shared modelsVending-Bench 2 and SWE-Bench verified: rho 0.60, 23 shared modelsVending-Bench 2 and LMArena Search: rho 0.61, 26 shared modelsVending-Bench 2 and Bug Hunt Bench: rho 0.73, 27 shared modelsVending-Bench 2 and LMArena Search (style control): rho 0.74, 26 shared modelsVending-Bench 2Vending-Bench 2 and LMArena Document (style control): rho 0.73, 31 shared modelsVending-Bench 2 and LMArena Document: rho 0.70, 31 shared modelsVending-Bench 2 and Vals Index: rho 0.72, 13 shared modelsVending-Bench 2 and Terminal-Bench 4: rho 0.68, 13 shared modelsVending-Bench 2 and Code Migration: rho 0.72, 13 shared modelsVending-Bench 2 and LMArena Agent: rho 0.64, 25 shared modelsVending-Bench 2 and LMArena WebDev: rho 0.85, 59 shared modelsVending-Bench 2 and CursorBench: rho 0.54, 15 shared modelsVending-Bench 2 and EMB: rho 0.52, 13 shared modelsVending-Bench 2 and FrontierCode: rho 0.54, 29 shared modelsVending-Bench 2 and Vibe Code Bench: rho 0.42, 13 shared modelsVending-Bench 2 and Legal Research: rho 0.37, 13 shared modelsVending-Bench 2 and Tax Agent: rho 0.30, 13 shared modelsVending-Bench 2 and Finance Agent: rho 0.20, 13 shared modelsVending-Bench 2 and Legal Agent Benchmark (HLAB-Vals): rho 0.28, 13 shared modelsVending-Bench 2 and GDPval: not enough shared modelsVending-Bench 2 and LMArena Image Edit: not enough shared modelsVending-Bench 2 and LMArena Text-to-Image: not enough shared modelsLMArena Document (style control) and LMArena Text (style control): rho 0.79, 39 shared modelsLMArena Document (style control) and LMArena Text: rho 0.61, 39 shared modelsLMArena Document (style control) and LMArena Vision (style control): rho 0.78, 37 shared modelsLMArena Document (style control) and LMArena Vision: rho 0.69, 37 shared modelsLMArena Document (style control) and GPQA diamond: rho 0.41, 36 shared modelsLMArena Document (style control) and METR time horizon: not enough shared modelsLMArena Document (style control) and FrontierMath-2025-02-28-Private: rho 0.61, 18 shared modelsLMArena Document (style control) and HLE: rho 0.59, 14 shared modelsLMArena Document (style control) and ARC-AGI-2: rho 0.79, 29 shared modelsLMArena Document (style control) and Terminal Bench: rho 0.69, 15 shared modelsLMArena Document (style control) and SWE-Bench verified: rho 0.77, 14 shared modelsLMArena Document (style control) and LMArena Search: rho 0.72, 21 shared modelsLMArena Document (style control) and Bug Hunt Bench: rho 0.55, 16 shared modelsLMArena Document (style control) and LMArena Search (style control): rho 0.81, 21 shared modelsLMArena Document (style control) and Vending-Bench 2: rho 0.73, 31 shared modelsLMArena Document (style control)LMArena Document (style control) and LMArena Document: rho 0.92, 40 shared modelsLMArena Document (style control) and Vals Index: not enough shared modelsLMArena Document (style control) and Terminal-Bench 4: not enough shared modelsLMArena Document (style control) and Code Migration: not enough shared modelsLMArena Document (style control) and LMArena Agent: rho 0.72, 15 shared modelsLMArena Document (style control) and LMArena WebDev: rho 0.77, 36 shared modelsLMArena Document (style control) and CursorBench: not enough shared modelsLMArena Document (style control) and EMB: not enough shared modelsLMArena Document (style control) and FrontierCode: rho 0.45, 20 shared modelsLMArena Document (style control) and Vibe Code Bench: not enough shared modelsLMArena Document (style control) and Legal Research: not enough shared modelsLMArena Document (style control) and Tax Agent: not enough shared modelsLMArena Document (style control) and Finance Agent: not enough shared modelsLMArena Document (style control) and Legal Agent Benchmark (HLAB-Vals): not enough shared modelsLMArena Document (style control) and GDPval: not enough shared modelsLMArena Document (style control) and LMArena Image Edit: not enough shared modelsLMArena Document (style control) and LMArena Text-to-Image: not enough shared modelsLMArena Document and LMArena Text (style control): rho 0.72, 39 shared modelsLMArena Document and LMArena Text: rho 0.67, 39 shared modelsLMArena Document and LMArena Vision (style control): rho 0.73, 37 shared modelsLMArena Document and LMArena Vision: rho 0.76, 37 shared modelsLMArena Document and GPQA diamond: rho 0.48, 36 shared modelsLMArena Document and METR time horizon: not enough shared modelsLMArena Document and FrontierMath-2025-02-28-Private: rho 0.49, 18 shared modelsLMArena Document and HLE: rho 0.44, 14 shared modelsLMArena Document and ARC-AGI-2: rho 0.78, 29 shared modelsLMArena Document and Terminal Bench: rho 0.55, 15 shared modelsLMArena Document and SWE-Bench verified: rho 0.74, 14 shared modelsLMArena Document and LMArena Search: rho 0.70, 21 shared modelsLMArena Document and Bug Hunt Bench: rho 0.45, 16 shared modelsLMArena Document and LMArena Search (style control): rho 0.74, 21 shared modelsLMArena Document and Vending-Bench 2: rho 0.70, 31 shared modelsLMArena Document and LMArena Document (style control): rho 0.92, 40 shared modelsLMArena DocumentLMArena Document and Vals Index: not enough shared modelsLMArena Document and Terminal-Bench 4: not enough shared modelsLMArena Document and Code Migration: not enough shared modelsLMArena Document and LMArena Agent: rho 0.79, 15 shared modelsLMArena Document and LMArena WebDev: rho 0.78, 36 shared modelsLMArena Document and CursorBench: not enough shared modelsLMArena Document and EMB: not enough shared modelsLMArena Document and FrontierCode: rho 0.39, 20 shared modelsLMArena Document and Vibe Code Bench: not enough shared modelsLMArena Document and Legal Research: not enough shared modelsLMArena Document and Tax Agent: not enough shared modelsLMArena Document and Finance Agent: not enough shared modelsLMArena Document and Legal Agent Benchmark (HLAB-Vals): not enough shared modelsLMArena Document and GDPval: not enough shared modelsLMArena Document and LMArena Image Edit: not enough shared modelsLMArena Document and LMArena Text-to-Image: not enough shared modelsVals Index and LMArena Text (style control): rho 0.63, 24 shared modelsVals Index and LMArena Text: rho 0.54, 24 shared modelsVals Index and LMArena Vision (style control): rho 0.64, 12 shared modelsVals Index and LMArena Vision: rho 0.43, 12 shared modelsVals Index and GPQA diamond: rho 0.85, 22 shared modelsVals Index and METR time horizon: not enough shared modelsVals Index and FrontierMath-2025-02-28-Private: not enough shared modelsVals Index and HLE: not enough shared modelsVals Index and ARC-AGI-2: rho 0.86, 20 shared modelsVals Index and Terminal Bench: not enough shared modelsVals Index and SWE-Bench verified: not enough shared modelsVals Index and LMArena Search: not enough shared modelsVals Index and Bug Hunt Bench: rho 0.87, 23 shared modelsVals Index and LMArena Search (style control): not enough shared modelsVals Index and Vending-Bench 2: rho 0.72, 13 shared modelsVals Index and LMArena Document (style control): not enough shared modelsVals Index and LMArena Document: not enough shared modelsVals IndexVals Index and Terminal-Bench 4: rho 0.94, 38 shared modelsVals Index and Code Migration: rho 0.93, 38 shared modelsVals Index and LMArena Agent: rho 0.90, 31 shared modelsVals Index and LMArena WebDev: rho 0.89, 32 shared modelsVals Index and CursorBench: rho 0.89, 12 shared modelsVals Index and EMB: rho 0.84, 38 shared modelsVals Index and FrontierCode: rho 0.82, 14 shared modelsVals Index and Vibe Code Bench: rho 0.85, 38 shared modelsVals Index and Legal Research: rho 0.68, 38 shared modelsVals Index and Tax Agent: rho 0.65, 38 shared modelsVals Index and Finance Agent: rho 0.77, 38 shared modelsVals Index and Legal Agent Benchmark (HLAB-Vals): rho 0.34, 38 shared modelsVals Index and GDPval: not enough shared modelsVals Index and LMArena Image Edit: not enough shared modelsVals Index and LMArena Text-to-Image: not enough shared modelsTerminal-Bench 4 and LMArena Text (style control): rho 0.68, 24 shared modelsTerminal-Bench 4 and LMArena Text: rho 0.56, 24 shared modelsTerminal-Bench 4 and LMArena Vision (style control): rho 0.64, 12 shared modelsTerminal-Bench 4 and LMArena Vision: rho 0.64, 12 shared modelsTerminal-Bench 4 and GPQA diamond: rho 0.76, 22 shared modelsTerminal-Bench 4 and METR time horizon: not enough shared modelsTerminal-Bench 4 and FrontierMath-2025-02-28-Private: not enough shared modelsTerminal-Bench 4 and HLE: not enough shared modelsTerminal-Bench 4 and ARC-AGI-2: rho 0.85, 20 shared modelsTerminal-Bench 4 and Terminal Bench: not enough shared modelsTerminal-Bench 4 and SWE-Bench verified: not enough shared modelsTerminal-Bench 4 and LMArena Search: not enough shared modelsTerminal-Bench 4 and Bug Hunt Bench: rho 0.80, 23 shared modelsTerminal-Bench 4 and LMArena Search (style control): not enough shared modelsTerminal-Bench 4 and Vending-Bench 2: rho 0.68, 13 shared modelsTerminal-Bench 4 and LMArena Document (style control): not enough shared modelsTerminal-Bench 4 and LMArena Document: not enough shared modelsTerminal-Bench 4 and Vals Index: rho 0.94, 38 shared modelsTerminal-Bench 4Terminal-Bench 4 and Code Migration: rho 0.92, 38 shared modelsTerminal-Bench 4 and LMArena Agent: rho 0.82, 31 shared modelsTerminal-Bench 4 and LMArena WebDev: rho 0.85, 32 shared modelsTerminal-Bench 4 and CursorBench: rho 0.81, 12 shared modelsTerminal-Bench 4 and EMB: rho 0.80, 38 shared modelsTerminal-Bench 4 and FrontierCode: rho 0.78, 14 shared modelsTerminal-Bench 4 and Vibe Code Bench: rho 0.77, 38 shared modelsTerminal-Bench 4 and Legal Research: rho 0.59, 38 shared modelsTerminal-Bench 4 and Tax Agent: rho 0.53, 38 shared modelsTerminal-Bench 4 and Finance Agent: rho 0.66, 38 shared modelsTerminal-Bench 4 and Legal Agent Benchmark (HLAB-Vals): rho 0.24, 38 shared modelsTerminal-Bench 4 and GDPval: not enough shared modelsTerminal-Bench 4 and LMArena Image Edit: not enough shared modelsTerminal-Bench 4 and LMArena Text-to-Image: not enough shared modelsCode Migration and LMArena Text (style control): rho 0.59, 24 shared modelsCode Migration and LMArena Text: rho 0.48, 24 shared modelsCode Migration and LMArena Vision (style control): rho 0.55, 12 shared modelsCode Migration and LMArena Vision: rho 0.42, 12 shared modelsCode Migration and GPQA diamond: rho 0.80, 22 shared modelsCode Migration and METR time horizon: not enough shared modelsCode Migration and FrontierMath-2025-02-28-Private: not enough shared modelsCode Migration and HLE: not enough shared modelsCode Migration and ARC-AGI-2: rho 0.83, 20 shared modelsCode Migration and Terminal Bench: not enough shared modelsCode Migration and SWE-Bench verified: not enough shared modelsCode Migration and LMArena Search: not enough shared modelsCode Migration and Bug Hunt Bench: rho 0.84, 23 shared modelsCode Migration and LMArena Search (style control): not enough shared modelsCode Migration and Vending-Bench 2: rho 0.72, 13 shared modelsCode Migration and LMArena Document (style control): not enough shared modelsCode Migration and LMArena Document: not enough shared modelsCode Migration and Vals Index: rho 0.93, 38 shared modelsCode Migration and Terminal-Bench 4: rho 0.92, 38 shared modelsCode MigrationCode Migration and LMArena Agent: rho 0.84, 31 shared modelsCode Migration and LMArena WebDev: rho 0.84, 32 shared modelsCode Migration and CursorBench: rho 0.81, 12 shared modelsCode Migration and EMB: rho 0.76, 38 shared modelsCode Migration and FrontierCode: rho 0.72, 14 shared modelsCode Migration and Vibe Code Bench: rho 0.79, 38 shared modelsCode Migration and Legal Research: rho 0.57, 38 shared modelsCode Migration and Tax Agent: rho 0.55, 38 shared modelsCode Migration and Finance Agent: rho 0.59, 38 shared modelsCode Migration and Legal Agent Benchmark (HLAB-Vals): rho 0.18, 38 shared modelsCode Migration and GDPval: not enough shared modelsCode Migration and LMArena Image Edit: not enough shared modelsCode Migration and LMArena Text-to-Image: not enough shared modelsLMArena Agent and LMArena Text (style control): rho 0.58, 46 shared modelsLMArena Agent and LMArena Text: rho 0.47, 46 shared modelsLMArena Agent and LMArena Vision (style control): rho 0.51, 32 shared modelsLMArena Agent and LMArena Vision: rho 0.42, 32 shared modelsLMArena Agent and GPQA diamond: rho 0.77, 35 shared modelsLMArena Agent and METR time horizon: not enough shared modelsLMArena Agent and FrontierMath-2025-02-28-Private: not enough shared modelsLMArena Agent and HLE: not enough shared modelsLMArena Agent and ARC-AGI-2: rho 0.82, 29 shared modelsLMArena Agent and Terminal Bench: not enough shared modelsLMArena Agent and SWE-Bench verified: not enough shared modelsLMArena Agent and LMArena Search: rho 0.08, 12 shared modelsLMArena Agent and Bug Hunt Bench: rho 0.80, 37 shared modelsLMArena Agent and LMArena Search (style control): rho 0.37, 12 shared modelsLMArena Agent and Vending-Bench 2: rho 0.64, 25 shared modelsLMArena Agent and LMArena Document (style control): rho 0.72, 15 shared modelsLMArena Agent and LMArena Document: rho 0.79, 15 shared modelsLMArena Agent and Vals Index: rho 0.90, 31 shared modelsLMArena Agent and Terminal-Bench 4: rho 0.82, 31 shared modelsLMArena Agent and Code Migration: rho 0.84, 31 shared modelsLMArena AgentLMArena Agent and LMArena WebDev: rho 0.90, 55 shared modelsLMArena Agent and CursorBench: rho 0.81, 19 shared modelsLMArena Agent and EMB: rho 0.83, 31 shared modelsLMArena Agent and FrontierCode: rho 0.87, 27 shared modelsLMArena Agent and Vibe Code Bench: rho 0.81, 31 shared modelsLMArena Agent and Legal Research: rho 0.52, 31 shared modelsLMArena Agent and Tax Agent: rho 0.49, 31 shared modelsLMArena Agent and Finance Agent: rho 0.64, 31 shared modelsLMArena Agent and Legal Agent Benchmark (HLAB-Vals): rho 0.18, 31 shared modelsLMArena Agent and GDPval: not enough shared modelsLMArena Agent and LMArena Image Edit: not enough shared modelsLMArena Agent and LMArena Text-to-Image: not enough shared modelsLMArena WebDev and LMArena Text (style control): rho 0.73, 112 shared modelsLMArena WebDev and LMArena Text: rho 0.65, 112 shared modelsLMArena WebDev and LMArena Vision (style control): rho 0.59, 63 shared modelsLMArena WebDev and LMArena Vision: rho 0.54, 63 shared modelsLMArena WebDev and GPQA diamond: rho 0.82, 78 shared modelsLMArena WebDev and METR time horizon: not enough shared modelsLMArena WebDev and FrontierMath-2025-02-28-Private: rho 0.72, 28 shared modelsLMArena WebDev and HLE: rho 0.82, 18 shared modelsLMArena WebDev and ARC-AGI-2: rho 0.84, 58 shared modelsLMArena WebDev and Terminal Bench: rho 0.71, 30 shared modelsLMArena WebDev and SWE-Bench verified: rho 0.78, 24 shared modelsLMArena WebDev and LMArena Search: rho 0.69, 31 shared modelsLMArena WebDev and Bug Hunt Bench: rho 0.81, 44 shared modelsLMArena WebDev and LMArena Search (style control): rho 0.84, 31 shared modelsLMArena WebDev and Vending-Bench 2: rho 0.85, 59 shared modelsLMArena WebDev and LMArena Document (style control): rho 0.77, 36 shared modelsLMArena WebDev and LMArena Document: rho 0.78, 36 shared modelsLMArena WebDev and Vals Index: rho 0.89, 32 shared modelsLMArena WebDev and Terminal-Bench 4: rho 0.85, 32 shared modelsLMArena WebDev and Code Migration: rho 0.84, 32 shared modelsLMArena WebDev and LMArena Agent: rho 0.90, 55 shared modelsLMArena WebDevLMArena WebDev and CursorBench: rho 0.88, 20 shared modelsLMArena WebDev and EMB: rho 0.82, 32 shared modelsLMArena WebDev and FrontierCode: rho 0.81, 38 shared modelsLMArena WebDev and Vibe Code Bench: rho 0.82, 32 shared modelsLMArena WebDev and Legal Research: rho 0.55, 32 shared modelsLMArena WebDev and Tax Agent: rho 0.51, 32 shared modelsLMArena WebDev and Finance Agent: rho 0.64, 32 shared modelsLMArena WebDev and Legal Agent Benchmark (HLAB-Vals): rho 0.14, 32 shared modelsLMArena WebDev and GDPval: not enough shared modelsLMArena WebDev and LMArena Image Edit: not enough shared modelsLMArena WebDev and LMArena Text-to-Image: not enough shared modelsCursorBench and LMArena Text (style control): rho 0.62, 18 shared modelsCursorBench and LMArena Text: rho 0.65, 18 shared modelsCursorBench and LMArena Vision (style control): rho 0.47, 12 shared modelsCursorBench and LMArena Vision: rho 0.57, 12 shared modelsCursorBench and GPQA diamond: rho 0.69, 16 shared modelsCursorBench and METR time horizon: not enough shared modelsCursorBench and FrontierMath-2025-02-28-Private: not enough shared modelsCursorBench and HLE: not enough shared modelsCursorBench and ARC-AGI-2: rho 0.69, 15 shared modelsCursorBench and Terminal Bench: not enough shared modelsCursorBench and SWE-Bench verified: not enough shared modelsCursorBench and LMArena Search: not enough shared modelsCursorBench and Bug Hunt Bench: rho 0.69, 20 shared modelsCursorBench and LMArena Search (style control): not enough shared modelsCursorBench and Vending-Bench 2: rho 0.54, 15 shared modelsCursorBench and LMArena Document (style control): not enough shared modelsCursorBench and LMArena Document: not enough shared modelsCursorBench and Vals Index: rho 0.89, 12 shared modelsCursorBench and Terminal-Bench 4: rho 0.81, 12 shared modelsCursorBench and Code Migration: rho 0.81, 12 shared modelsCursorBench and LMArena Agent: rho 0.81, 19 shared modelsCursorBench and LMArena WebDev: rho 0.88, 20 shared modelsCursorBenchCursorBench and EMB: rho 0.89, 12 shared modelsCursorBench and FrontierCode: rho 0.80, 15 shared modelsCursorBench and Vibe Code Bench: rho 0.70, 12 shared modelsCursorBench and Legal Research: rho 0.56, 12 shared modelsCursorBench and Tax Agent: rho 0.52, 12 shared modelsCursorBench and Finance Agent: rho 0.79, 12 shared modelsCursorBench and Legal Agent Benchmark (HLAB-Vals): rho -0.15, 12 shared modelsCursorBench and GDPval: not enough shared modelsCursorBench and LMArena Image Edit: not enough shared modelsCursorBench and LMArena Text-to-Image: not enough shared modelsEMB and LMArena Text (style control): rho 0.72, 24 shared modelsEMB and LMArena Text: rho 0.58, 24 shared modelsEMB and LMArena Vision (style control): rho 0.66, 12 shared modelsEMB and LMArena Vision: rho 0.43, 12 shared modelsEMB and GPQA diamond: rho 0.90, 22 shared modelsEMB and METR time horizon: not enough shared modelsEMB and FrontierMath-2025-02-28-Private: not enough shared modelsEMB and HLE: not enough shared modelsEMB and ARC-AGI-2: rho 0.81, 20 shared modelsEMB and Terminal Bench: not enough shared modelsEMB and SWE-Bench verified: not enough shared modelsEMB and LMArena Search: not enough shared modelsEMB and Bug Hunt Bench: rho 0.69, 23 shared modelsEMB and LMArena Search (style control): not enough shared modelsEMB and Vending-Bench 2: rho 0.52, 13 shared modelsEMB and LMArena Document (style control): not enough shared modelsEMB and LMArena Document: not enough shared modelsEMB and Vals Index: rho 0.84, 38 shared modelsEMB and Terminal-Bench 4: rho 0.80, 38 shared modelsEMB and Code Migration: rho 0.76, 38 shared modelsEMB and LMArena Agent: rho 0.83, 31 shared modelsEMB and LMArena WebDev: rho 0.82, 32 shared modelsEMB and CursorBench: rho 0.89, 12 shared modelsEMBEMB and FrontierCode: rho 0.77, 14 shared modelsEMB and Vibe Code Bench: rho 0.73, 38 shared modelsEMB and Legal Research: rho 0.64, 38 shared modelsEMB and Tax Agent: rho 0.60, 38 shared modelsEMB and Finance Agent: rho 0.64, 38 shared modelsEMB and Legal Agent Benchmark (HLAB-Vals): rho 0.30, 38 shared modelsEMB and GDPval: not enough shared modelsEMB and LMArena Image Edit: not enough shared modelsEMB and LMArena Text-to-Image: not enough shared modelsFrontierCode and LMArena Text (style control): rho 0.39, 38 shared modelsFrontierCode and LMArena Text: rho 0.17, 38 shared modelsFrontierCode and LMArena Vision (style control): rho 0.33, 30 shared modelsFrontierCode and LMArena Vision: rho 0.26, 30 shared modelsFrontierCode and GPQA diamond: rho 0.68, 40 shared modelsFrontierCode and METR time horizon: not enough shared modelsFrontierCode and FrontierMath-2025-02-28-Private: not enough shared modelsFrontierCode and HLE: not enough shared modelsFrontierCode and ARC-AGI-2: rho 0.71, 33 shared modelsFrontierCode and Terminal Bench: not enough shared modelsFrontierCode and SWE-Bench verified: not enough shared modelsFrontierCode and LMArena Search: rho 0.10, 16 shared modelsFrontierCode and Bug Hunt Bench: rho 0.69, 28 shared modelsFrontierCode and LMArena Search (style control): rho 0.51, 16 shared modelsFrontierCode and Vending-Bench 2: rho 0.54, 29 shared modelsFrontierCode and LMArena Document (style control): rho 0.45, 20 shared modelsFrontierCode and LMArena Document: rho 0.39, 20 shared modelsFrontierCode and Vals Index: rho 0.82, 14 shared modelsFrontierCode and Terminal-Bench 4: rho 0.78, 14 shared modelsFrontierCode and Code Migration: rho 0.72, 14 shared modelsFrontierCode and LMArena Agent: rho 0.87, 27 shared modelsFrontierCode and LMArena WebDev: rho 0.81, 38 shared modelsFrontierCode and CursorBench: rho 0.80, 15 shared modelsFrontierCode and EMB: rho 0.77, 14 shared modelsFrontierCodeFrontierCode and Vibe Code Bench: rho 0.82, 14 shared modelsFrontierCode and Legal Research: rho 0.38, 14 shared modelsFrontierCode and Tax Agent: rho 0.37, 14 shared modelsFrontierCode and Finance Agent: rho 0.53, 14 shared modelsFrontierCode and Legal Agent Benchmark (HLAB-Vals): rho 0.01, 14 shared modelsFrontierCode and GDPval: not enough shared modelsFrontierCode and LMArena Image Edit: not enough shared modelsFrontierCode and LMArena Text-to-Image: not enough shared modelsVibe Code Bench and LMArena Text (style control): rho 0.42, 24 shared modelsVibe Code Bench and LMArena Text: rho 0.29, 24 shared modelsVibe Code Bench and LMArena Vision (style control): rho 0.50, 12 shared modelsVibe Code Bench and LMArena Vision: rho 0.27, 12 shared modelsVibe Code Bench and GPQA diamond: rho 0.72, 22 shared modelsVibe Code Bench and METR time horizon: not enough shared modelsVibe Code Bench and FrontierMath-2025-02-28-Private: not enough shared modelsVibe Code Bench and HLE: not enough shared modelsVibe Code Bench and ARC-AGI-2: rho 0.70, 20 shared modelsVibe Code Bench and Terminal Bench: not enough shared modelsVibe Code Bench and SWE-Bench verified: not enough shared modelsVibe Code Bench and LMArena Search: not enough shared modelsVibe Code Bench and Bug Hunt Bench: rho 0.78, 23 shared modelsVibe Code Bench and LMArena Search (style control): not enough shared modelsVibe Code Bench and Vending-Bench 2: rho 0.42, 13 shared modelsVibe Code Bench and LMArena Document (style control): not enough shared modelsVibe Code Bench and LMArena Document: not enough shared modelsVibe Code Bench and Vals Index: rho 0.85, 38 shared modelsVibe Code Bench and Terminal-Bench 4: rho 0.77, 38 shared modelsVibe Code Bench and Code Migration: rho 0.79, 38 shared modelsVibe Code Bench and LMArena Agent: rho 0.81, 31 shared modelsVibe Code Bench and LMArena WebDev: rho 0.82, 32 shared modelsVibe Code Bench and CursorBench: rho 0.70, 12 shared modelsVibe Code Bench and EMB: rho 0.73, 38 shared modelsVibe Code Bench and FrontierCode: rho 0.82, 14 shared modelsVibe Code BenchVibe Code Bench and Legal Research: rho 0.59, 38 shared modelsVibe Code Bench and Tax Agent: rho 0.50, 38 shared modelsVibe Code Bench and Finance Agent: rho 0.68, 38 shared modelsVibe Code Bench and Legal Agent Benchmark (HLAB-Vals): rho 0.21, 38 shared modelsVibe Code Bench and GDPval: not enough shared modelsVibe Code Bench and LMArena Image Edit: not enough shared modelsVibe Code Bench and LMArena Text-to-Image: not enough shared modelsLegal Research and LMArena Text (style control): rho 0.65, 24 shared modelsLegal Research and LMArena Text: rho 0.68, 24 shared modelsLegal Research and LMArena Vision (style control): rho 0.44, 12 shared modelsLegal Research and LMArena Vision: rho 0.46, 12 shared modelsLegal Research and GPQA diamond: rho 0.50, 22 shared modelsLegal Research and METR time horizon: not enough shared modelsLegal Research and FrontierMath-2025-02-28-Private: not enough shared modelsLegal Research and HLE: not enough shared modelsLegal Research and ARC-AGI-2: rho 0.59, 20 shared modelsLegal Research and Terminal Bench: not enough shared modelsLegal Research and SWE-Bench verified: not enough shared modelsLegal Research and LMArena Search: not enough shared modelsLegal Research and Bug Hunt Bench: rho 0.27, 23 shared modelsLegal Research and LMArena Search (style control): not enough shared modelsLegal Research and Vending-Bench 2: rho 0.37, 13 shared modelsLegal Research and LMArena Document (style control): not enough shared modelsLegal Research and LMArena Document: not enough shared modelsLegal Research and Vals Index: rho 0.68, 38 shared modelsLegal Research and Terminal-Bench 4: rho 0.59, 38 shared modelsLegal Research and Code Migration: rho 0.57, 38 shared modelsLegal Research and LMArena Agent: rho 0.52, 31 shared modelsLegal Research and LMArena WebDev: rho 0.55, 32 shared modelsLegal Research and CursorBench: rho 0.56, 12 shared modelsLegal Research and EMB: rho 0.64, 38 shared modelsLegal Research and FrontierCode: rho 0.38, 14 shared modelsLegal Research and Vibe Code Bench: rho 0.59, 38 shared modelsLegal ResearchLegal Research and Tax Agent: rho 0.89, 38 shared modelsLegal Research and Finance Agent: rho 0.76, 38 shared modelsLegal Research and Legal Agent Benchmark (HLAB-Vals): rho 0.64, 38 shared modelsLegal Research and GDPval: not enough shared modelsLegal Research and LMArena Image Edit: not enough shared modelsLegal Research and LMArena Text-to-Image: not enough shared modelsTax Agent and LMArena Text (style control): rho 0.62, 24 shared modelsTax Agent and LMArena Text: rho 0.68, 24 shared modelsTax Agent and LMArena Vision (style control): rho 0.50, 12 shared modelsTax Agent and LMArena Vision: rho 0.48, 12 shared modelsTax Agent and GPQA diamond: rho 0.46, 22 shared modelsTax Agent and METR time horizon: not enough shared modelsTax Agent and FrontierMath-2025-02-28-Private: not enough shared modelsTax Agent and HLE: not enough shared modelsTax Agent and ARC-AGI-2: rho 0.56, 20 shared modelsTax Agent and Terminal Bench: not enough shared modelsTax Agent and SWE-Bench verified: not enough shared modelsTax Agent and LMArena Search: not enough shared modelsTax Agent and Bug Hunt Bench: rho 0.20, 23 shared modelsTax Agent and LMArena Search (style control): not enough shared modelsTax Agent and Vending-Bench 2: rho 0.30, 13 shared modelsTax Agent and LMArena Document (style control): not enough shared modelsTax Agent and LMArena Document: not enough shared modelsTax Agent and Vals Index: rho 0.65, 38 shared modelsTax Agent and Terminal-Bench 4: rho 0.53, 38 shared modelsTax Agent and Code Migration: rho 0.55, 38 shared modelsTax Agent and LMArena Agent: rho 0.49, 31 shared modelsTax Agent and LMArena WebDev: rho 0.51, 32 shared modelsTax Agent and CursorBench: rho 0.52, 12 shared modelsTax Agent and EMB: rho 0.60, 38 shared modelsTax Agent and FrontierCode: rho 0.37, 14 shared modelsTax Agent and Vibe Code Bench: rho 0.50, 38 shared modelsTax Agent and Legal Research: rho 0.89, 38 shared modelsTax AgentTax Agent and Finance Agent: rho 0.73, 38 shared modelsTax Agent and Legal Agent Benchmark (HLAB-Vals): rho 0.66, 38 shared modelsTax Agent and GDPval: not enough shared modelsTax Agent and LMArena Image Edit: not enough shared modelsTax Agent and LMArena Text-to-Image: not enough shared modelsFinance Agent and LMArena Text (style control): rho 0.56, 24 shared modelsFinance Agent and LMArena Text: rho 0.60, 24 shared modelsFinance Agent and LMArena Vision (style control): rho 0.69, 12 shared modelsFinance Agent and LMArena Vision: rho 0.71, 12 shared modelsFinance Agent and GPQA diamond: rho 0.57, 22 shared modelsFinance Agent and METR time horizon: not enough shared modelsFinance Agent and FrontierMath-2025-02-28-Private: not enough shared modelsFinance Agent and HLE: not enough shared modelsFinance Agent and ARC-AGI-2: rho 0.71, 20 shared modelsFinance Agent and Terminal Bench: not enough shared modelsFinance Agent and SWE-Bench verified: not enough shared modelsFinance Agent and LMArena Search: not enough shared modelsFinance Agent and Bug Hunt Bench: rho 0.39, 23 shared modelsFinance Agent and LMArena Search (style control): not enough shared modelsFinance Agent and Vending-Bench 2: rho 0.20, 13 shared modelsFinance Agent and LMArena Document (style control): not enough shared modelsFinance Agent and LMArena Document: not enough shared modelsFinance Agent and Vals Index: rho 0.77, 38 shared modelsFinance Agent and Terminal-Bench 4: rho 0.66, 38 shared modelsFinance Agent and Code Migration: rho 0.59, 38 shared modelsFinance Agent and LMArena Agent: rho 0.64, 31 shared modelsFinance Agent and LMArena WebDev: rho 0.64, 32 shared modelsFinance Agent and CursorBench: rho 0.79, 12 shared modelsFinance Agent and EMB: rho 0.64, 38 shared modelsFinance Agent and FrontierCode: rho 0.53, 14 shared modelsFinance Agent and Vibe Code Bench: rho 0.68, 38 shared modelsFinance Agent and Legal Research: rho 0.76, 38 shared modelsFinance Agent and Tax Agent: rho 0.73, 38 shared modelsFinance AgentFinance Agent and Legal Agent Benchmark (HLAB-Vals): rho 0.53, 38 shared modelsFinance Agent and GDPval: not enough shared modelsFinance Agent and LMArena Image Edit: not enough shared modelsFinance Agent and LMArena Text-to-Image: not enough shared modelsLegal Agent Benchmark (HLAB-Vals) and LMArena Text (style control): rho 0.49, 24 shared modelsLegal Agent Benchmark (HLAB-Vals) and LMArena Text: rho 0.50, 24 shared modelsLegal Agent Benchmark (HLAB-Vals) and LMArena Vision (style control): rho 0.02, 12 shared modelsLegal Agent Benchmark (HLAB-Vals) and LMArena Vision: rho -0.05, 12 shared modelsLegal Agent Benchmark (HLAB-Vals) and GPQA diamond: rho 0.19, 22 shared modelsLegal Agent Benchmark (HLAB-Vals) and METR time horizon: not enough shared modelsLegal Agent Benchmark (HLAB-Vals) and FrontierMath-2025-02-28-Private: not enough shared modelsLegal Agent Benchmark (HLAB-Vals) and HLE: not enough shared modelsLegal Agent Benchmark (HLAB-Vals) and ARC-AGI-2: rho 0.29, 20 shared modelsLegal Agent Benchmark (HLAB-Vals) and Terminal Bench: not enough shared modelsLegal Agent Benchmark (HLAB-Vals) and SWE-Bench verified: not enough shared modelsLegal Agent Benchmark (HLAB-Vals) and LMArena Search: not enough shared modelsLegal Agent Benchmark (HLAB-Vals) and Bug Hunt Bench: rho 0.01, 23 shared modelsLegal Agent Benchmark (HLAB-Vals) and LMArena Search (style control): not enough shared modelsLegal Agent Benchmark (HLAB-Vals) and Vending-Bench 2: rho 0.28, 13 shared modelsLegal Agent Benchmark (HLAB-Vals) and LMArena Document (style control): not enough shared modelsLegal Agent Benchmark (HLAB-Vals) and LMArena Document: not enough shared modelsLegal Agent Benchmark (HLAB-Vals) and Vals Index: rho 0.34, 38 shared modelsLegal Agent Benchmark (HLAB-Vals) and Terminal-Bench 4: rho 0.24, 38 shared modelsLegal Agent Benchmark (HLAB-Vals) and Code Migration: rho 0.18, 38 shared modelsLegal Agent Benchmark (HLAB-Vals) and LMArena Agent: rho 0.18, 31 shared modelsLegal Agent Benchmark (HLAB-Vals) and LMArena WebDev: rho 0.14, 32 shared modelsLegal Agent Benchmark (HLAB-Vals) and CursorBench: rho -0.15, 12 shared modelsLegal Agent Benchmark (HLAB-Vals) and EMB: rho 0.30, 38 shared modelsLegal Agent Benchmark (HLAB-Vals) and FrontierCode: rho 0.01, 14 shared modelsLegal Agent Benchmark (HLAB-Vals) and Vibe Code Bench: rho 0.21, 38 shared modelsLegal Agent Benchmark (HLAB-Vals) and Legal Research: rho 0.64, 38 shared modelsLegal Agent Benchmark (HLAB-Vals) and Tax Agent: rho 0.66, 38 shared modelsLegal Agent Benchmark (HLAB-Vals) and Finance Agent: rho 0.53, 38 shared modelsLegal Agent Benchmark (HLAB-Vals)Legal Agent Benchmark (HLAB-Vals) and GDPval: not enough shared modelsLegal Agent Benchmark (HLAB-Vals) and LMArena Image Edit: not enough shared modelsLegal Agent Benchmark (HLAB-Vals) and LMArena Text-to-Image: not enough shared modelsGDPval and LMArena Text (style control): rho 0.60, 10 shared modelsGDPval and LMArena Text: rho 0.28, 10 shared modelsGDPval and LMArena Vision (style control): not enough shared modelsGDPval and LMArena Vision: not enough shared modelsGDPval and GPQA diamond: rho 0.33, 11 shared modelsGDPval and METR time horizon: not enough shared modelsGDPval and FrontierMath-2025-02-28-Private: rho 0.44, 11 shared modelsGDPval and HLE: not enough shared modelsGDPval and ARC-AGI-2: rho 0.79, 10 shared modelsGDPval and Terminal Bench: not enough shared modelsGDPval and SWE-Bench verified: not enough shared modelsGDPval and LMArena Search: not enough shared modelsGDPval and Bug Hunt Bench: not enough shared modelsGDPval and LMArena Search (style control): not enough shared modelsGDPval and Vending-Bench 2: not enough shared modelsGDPval and LMArena Document (style control): not enough shared modelsGDPval and LMArena Document: not enough shared modelsGDPval and Vals Index: not enough shared modelsGDPval and Terminal-Bench 4: not enough shared modelsGDPval and Code Migration: not enough shared modelsGDPval and LMArena Agent: not enough shared modelsGDPval and LMArena WebDev: not enough shared modelsGDPval and CursorBench: not enough shared modelsGDPval and EMB: not enough shared modelsGDPval and FrontierCode: not enough shared modelsGDPval and Vibe Code Bench: not enough shared modelsGDPval and Legal Research: not enough shared modelsGDPval and Tax Agent: not enough shared modelsGDPval and Finance Agent: not enough shared modelsGDPval and Legal Agent Benchmark (HLAB-Vals): not enough shared modelsGDPvalGDPval and LMArena Image Edit: not enough shared modelsGDPval and LMArena Text-to-Image: not enough shared modelsLMArena Image Edit and LMArena Text (style control): not enough shared modelsLMArena Image Edit and LMArena Text: not enough shared modelsLMArena Image Edit and LMArena Vision (style control): not enough shared modelsLMArena Image Edit and LMArena Vision: not enough shared modelsLMArena Image Edit and GPQA diamond: not enough shared modelsLMArena Image Edit and METR time horizon: not enough shared modelsLMArena Image Edit and FrontierMath-2025-02-28-Private: not enough shared modelsLMArena Image Edit and HLE: not enough shared modelsLMArena Image Edit and ARC-AGI-2: not enough shared modelsLMArena Image Edit and Terminal Bench: not enough shared modelsLMArena Image Edit and SWE-Bench verified: not enough shared modelsLMArena Image Edit and LMArena Search: not enough shared modelsLMArena Image Edit and Bug Hunt Bench: not enough shared modelsLMArena Image Edit and LMArena Search (style control): not enough shared modelsLMArena Image Edit and Vending-Bench 2: not enough shared modelsLMArena Image Edit and LMArena Document (style control): not enough shared modelsLMArena Image Edit and LMArena Document: not enough shared modelsLMArena Image Edit and Vals Index: not enough shared modelsLMArena Image Edit and Terminal-Bench 4: not enough shared modelsLMArena Image Edit and Code Migration: not enough shared modelsLMArena Image Edit and LMArena Agent: not enough shared modelsLMArena Image Edit and LMArena WebDev: not enough shared modelsLMArena Image Edit and CursorBench: not enough shared modelsLMArena Image Edit and EMB: not enough shared modelsLMArena Image Edit and FrontierCode: not enough shared modelsLMArena Image Edit and Vibe Code Bench: not enough shared modelsLMArena Image Edit and Legal Research: not enough shared modelsLMArena Image Edit and Tax Agent: not enough shared modelsLMArena Image Edit and Finance Agent: not enough shared modelsLMArena Image Edit and Legal Agent Benchmark (HLAB-Vals): not enough shared modelsLMArena Image Edit and GDPval: not enough shared modelsLMArena Image EditLMArena Image Edit and LMArena Text-to-Image: rho 0.93, 43 shared modelsLMArena Text-to-Image and LMArena Text (style control): not enough shared modelsLMArena Text-to-Image and LMArena Text: not enough shared modelsLMArena Text-to-Image and LMArena Vision (style control): not enough shared modelsLMArena Text-to-Image and LMArena Vision: not enough shared modelsLMArena Text-to-Image and GPQA diamond: not enough shared modelsLMArena Text-to-Image and METR time horizon: not enough shared modelsLMArena Text-to-Image and FrontierMath-2025-02-28-Private: not enough shared modelsLMArena Text-to-Image and HLE: not enough shared modelsLMArena Text-to-Image and ARC-AGI-2: not enough shared modelsLMArena Text-to-Image and Terminal Bench: not enough shared modelsLMArena Text-to-Image and SWE-Bench verified: not enough shared modelsLMArena Text-to-Image and LMArena Search: not enough shared modelsLMArena Text-to-Image and Bug Hunt Bench: not enough shared modelsLMArena Text-to-Image and LMArena Search (style control): not enough shared modelsLMArena Text-to-Image and Vending-Bench 2: not enough shared modelsLMArena Text-to-Image and LMArena Document (style control): not enough shared modelsLMArena Text-to-Image and LMArena Document: not enough shared modelsLMArena Text-to-Image and Vals Index: not enough shared modelsLMArena Text-to-Image and Terminal-Bench 4: not enough shared modelsLMArena Text-to-Image and Code Migration: not enough shared modelsLMArena Text-to-Image and LMArena Agent: not enough shared modelsLMArena Text-to-Image and LMArena WebDev: not enough shared modelsLMArena Text-to-Image and CursorBench: not enough shared modelsLMArena Text-to-Image and EMB: not enough shared modelsLMArena Text-to-Image and FrontierCode: not enough shared modelsLMArena Text-to-Image and Vibe Code Bench: not enough shared modelsLMArena Text-to-Image and Legal Research: not enough shared modelsLMArena Text-to-Image and Tax Agent: not enough shared modelsLMArena Text-to-Image and Finance Agent: not enough shared modelsLMArena Text-to-Image and Legal Agent Benchmark (HLAB-Vals): not enough shared modelsLMArena Text-to-Image and GDPval: not enough shared modelsLMArena Text-to-Image and LMArena Image Edit: rho 0.93, 43 shared modelsLMArena Text-to-ImageLMArena Text (style control)LMArena TextLMArena Vision (style control)LMArena VisionGPQA diamondMETR time horizonFrontierMath-2025-02-28-Priva…HLEARC-AGI-2Terminal BenchSWE-Bench verifiedLMArena SearchBug Hunt BenchLMArena Search (style control)Vending-Bench 2LMArena Document (style contr…LMArena DocumentVals IndexTerminal-Bench 4Code MigrationLMArena AgentLMArena WebDevCursorBenchEMBFrontierCodeVibe Code BenchLegal ResearchTax AgentFinance AgentLegal Agent Benchmark (HLAB-V…GDPvalLMArena Image EditLMArena Text-to-ImageLMArena Text (style control)LMArena TextLMArena Vision (style control)LMArena VisionGPQA diamondMETR time horizonFrontierMath-2025-02-28-Priva…HLEARC-AGI-2Terminal BenchSWE-Bench verifiedLMArena SearchBug Hunt BenchLMArena Search (style control)Vending-Bench 2LMArena Document (style contr…LMArena DocumentVals IndexTerminal-Bench 4Code MigrationLMArena AgentLMArena WebDevCursorBenchEMBFrontierCodeVibe Code BenchLegal ResearchTax AgentFinance AgentLegal Agent Benchmark (HLAB-V…GDPvalLMArena Image EditLMArena Text-to-Image123456

The boards as a 3-D map

Drag to rotate, pinch or hold Ctrl and scroll to zoom, arrow keys to turn, brackets to step through points, plus and minus to zoom, 0 to reset.

33 of 33 boards shown. Lines are the tree of closest pairs. The layout is approximate: stress 0.32 (lower is more faithful). The axes have no meaning. Points that are near each other have alike model rankings, not alike tasks.

Table of the 33 boards and their clusters
Benchmark boards with their cluster and nearest neighbor in the tree
BoardClusterNearest neighborRho
LMArena Text (style control)Cluster 1LMArena Text0.98
LMArena TextCluster 1LMArena Text (style control)0.98
LMArena Vision (style control)Cluster 1LMArena Vision0.98
LMArena VisionCluster 1LMArena Vision (style control)0.98
GPQA diamondCluster 1METR time horizon0.95
METR time horizonCluster 1GPQA diamond0.95
FrontierMath-2025-02-28-PrivateCluster 1GPQA diamond0.94
HLECluster 1GPQA diamond0.95
ARC-AGI-2Cluster 1GPQA diamond0.93
Terminal BenchCluster 1HLE0.91
SWE-Bench verifiedCluster 1ARC-AGI-20.91
LMArena SearchCluster 1METR time horizon0.92
Bug Hunt BenchCluster 2Vals Index0.87
LMArena Search (style control)Cluster 2ARC-AGI-20.87
Vending-Bench 2Cluster 2LMArena WebDev0.85
LMArena Document (style control)Cluster 3LMArena Document0.92
LMArena DocumentCluster 3LMArena Document (style control)0.92
Vals IndexCluster 4Terminal-Bench 40.94
Terminal-Bench 4Cluster 4Vals Index0.94
Code MigrationCluster 4Vals Index0.93
LMArena AgentCluster 4LMArena WebDev0.90
LMArena WebDevCluster 4LMArena Agent0.90
CursorBenchCluster 4EMB0.89
EMBCluster 4GPQA diamond0.90
FrontierCodeCluster 4LMArena Agent0.87
Vibe Code BenchCluster 4Vals Index0.85
Legal ResearchCluster 5Tax Agent0.89
Tax AgentCluster 5Legal Research0.89
Finance AgentCluster 5CursorBench0.79
Legal Agent Benchmark (HLAB-Vals)On its ownTax Agent0.66
GDPvalOn its ownARC-AGI-20.79
LMArena Image EditCluster 6LMArena Text-to-Image0.93
LMArena Text-to-ImageCluster 6LMArena Image Edit0.93
  1. 1 12 boards, mean rho 0.86, mostly general-text
  2. 2 3 boards, mean rho 0.78, mostly agentic
  3. 3 2 boards, mean rho 0.92, mostly general-text
  4. 4 9 boards, mean rho 0.83, mostly coding
  5. 5 3 boards, mean rho 0.80, mostly finance
  6. 6 2 boards, mean rho 0.93, mostly image-gen
Method, limits, and excluded inputs (33 boards, 3509 rows read)

Method rules

  • Distance between two boards is d = sqrt(2 (1 - rho)) from their Spearman rho: 0 for identical rankings, sqrt(2) for no relation, 2 for reversed.
  • A pair is observed when the agreement analysis computed it from at least 10 shared models.
  • A pair that is not observed gets the shortest path through observed pairs; if no path exists it gets sqrt(2), the distance for no relation (or the largest observed distance, if that is larger). These distances place and group boards but are drawn blank in the heat map and left out of stress.
  • Boards with no observed pair at all are left out.
  • Position: classical multidimensional scaling to 3 dimensions (top 3 positive eigenvectors of the double-centred squared distances, Jacobi eigen solver, no random start), scaled by one common factor into the unit cube.
  • Stress is Kruskal stress-1 on observed pairs: sqrt(sum (embedded - observed)^2 / sum observed^2). Lower is a more faithful picture.
  • Clusters: average-linkage (UPGMA) hierarchical clustering, cut at distance 0.77 (average rho about 0.7). The leaf order of the dendrogram orders the heat map.
  • Tree: minimum spanning tree over observed pairs only (Kruskal), shortest distance first.

Known limits

  • Rank agreement is not task identity: two boards can agree because strong models are strong at everything.
  • Pairwise correlations use different model subsets, so the distances are only approximately Euclidean; stress says how much is lost in 3 dimensions.
  • Imputed distances for unobserved pairs are a guess that keeps the layout connected; they carry no evidence.
  • Axes of the 3 dimensions have no meaning; only distances between boards do.

Exclusions

  • 16 boards excluded: board has fewer than 8 models
  • 94 boards excluded: not the default slice of its source and version
  • 1 boards excluded: older version of a benchmark the source reports again
  • 2 boards excluded: rank metric is not a quality score
  • 203 pairs excluded: board pair shares fewer than 6 models
  • 75 rows excluded: hidden row (for example superseded)
  • 1 boards excluded: board has no pair with at least 10 shared models

Near-redundant benchmark pairs

Board ABoard BShared modelsSpearman rhoKendall tau-b
lmarena-text-style-control/lmarena#overalllmarena-text/lmarena#overall3920.9840.899
lmarena-vision-style-control/lmarena#overalllmarena-vision/lmarena#overall1530.9830.893
lmarena-text-style-control/lmarena#overalllmarena-vision-style-control/lmarena#overall1250.9780.871
lmarena-text/lmarena#overalllmarena-vision/lmarena#overall1250.9680.860
lmarena-text-style-control/lmarena#overalllmarena-vision/lmarena#overall1250.9570.827
gpqa-diamond/epochmetr-time-horizon/metr@1.1200.9520.832
gpqa-diamond/epochhle/epoch360.9500.828
lmarena-text-style-control/lmarena#overallmetr-time-horizon/metr@1.1200.9500.821
lmarena-vision/lmarena#overallmetr-time-horizon/metr@1.1150.9500.829
lmarena-text/lmarena#overallmetr-time-horizon/metr@1.1200.9470.811
lmarena-vision-style-control/lmarena#overallmetr-time-horizon/metr@1.1150.9460.810
gpqa-diamond/epochlmarena-text-style-control/lmarena#overall1600.9430.792
vals-index/valsvals-terminal-bench-4/vals380.9420.821
frontiermath/epochgpqa-diamond/epoch670.9350.787
lmarena-text/lmarena#overalllmarena-vision-style-control/lmarena#overall1250.9350.786
frontiermath/epochhle/epoch260.9340.793
arc-agi-2/epochgpqa-diamond/epoch780.9310.785
lmarena-image-edit/lmarena#overalllmarena-text-to-image/lmarena#overall430.9300.790
vals-code-migration/valsvals-index/vals380.9300.784
arc-agi-2/epochhle/epoch310.9270.797
arc-agi-2/epochmetr-time-horizon/metr@1.1100.9270.822
gpqa-diamond/epochlmarena-text/lmarena#overall1600.9270.772
vals-code-migration/valsvals-terminal-bench-4/vals380.9240.779
frontiermath/epochlmarena-vision/lmarena#overall390.9180.759
lmarena-document-style-control/lmarenalmarena-document/lmarena400.9150.759
lmarena-search/lmarenametr-time-horizon/metr@1.1100.9150.778
frontiermath/epochmetr-time-horizon/metr@1.1140.9090.773
arc-agi-2/epochswe-bench-verified/epoch220.9070.761
hle/epochterminal-bench/epoch130.9050.761
lmarena-agent/lmarenalmarena-webdev/lmarena#overall550.9040.752
gpqa-diamond/epochvals-emb/vals220.9030.743
hle/epochmetr-time-horizon/metr@1.1130.9010.795

Within-domain agreement

DomainBoardsWithin-domain pairsMean Spearman rho
Agentic540.574
Coding9240.793
Finance430.655
General text5100.790
Image generation210.930
Legal210.641
Math10Insufficient pairs
Science210.950
Search210.811
Vision210.983
Method, limits, and excluded inputs (34 boards, 3509 rows read)

Method rules

  • Boards: one default quality board per benchmark and source with at least 8 models.
  • All metrics are oriented so higher values represent stronger capability before correlating.
  • Each board pair sharing at least 6 models computes Spearman rho and Kendall tau-b on each model's best visible row.
  • Pairs with rho >= 0.90 across at least 10 shared models are flagged as near-redundant.
  • Effective dimensionality N_eff = (sum lambda+)^2 / sum (lambda+^2) comes from the symmetric pairwise Spearman correlation matrix.

Known limits

  • Pairwise-complete correlations mix slightly different model subsets across benchmark pairs.
  • High rank correlation does not prove identical tasks; frontier models can improve simultaneously across distinct skills.

Exclusions

  • 16 boards excluded: board has fewer than 8 models
  • 94 boards excluded: not the default slice of its source and version
  • 1 boards excluded: older version of a benchmark the source reports again
  • 2 boards excluded: rank metric is not a quality score
  • 203 pairs excluded: board pair shares fewer than 6 models
  • 75 rows excluded: hidden row (for example superseded)