Grade A models
71
Model evidence grades (A to D), rank probability per board, #1 tie bands, single-run exposure, and vendor self-preference DiD checks.
71
124
63
Insufficient data
1 vendor-run boards
A board lists one rank per model, but each score carries error. We redraw every score 2,000 times within its 95% interval, rank the board each time, and count how often a model lands on each rank. Rows without an interval stay fixed, and errors are treated as independent.
Rows are listed best first on Accuracy (best score across scorers). Each cell is the share of draws in which the model holds that rank; the number is shown from 5%.
Method rules
Known limits
Exclusions
| Benchmark | Rows | Tied #1 count | Single-run rows |
|---|---|---|---|
| ARC-AGI-2Epoch AI Benchmarking Hub | 299 | Clear #1 | 0 |
| Bug Hunt BenchBug Hunt Bench | 145 | Clear #1 | 63 |
| CursorBenchCursorBench | 91 | Clear #1 | 0 |
| FrontierCodeEpoch AI Benchmarking Hub | 51 | Clear #1 | 0 |
| FrontierMath-2025-02-28-PrivateEpoch AI Benchmarking Hub | 101 | Tied #1 (6) | 0 |
| GDPvalEpoch AI Benchmarking Hub | 11 | Clear #1 | 0 |
| GPQA diamondEpoch AI Benchmarking Hub | 338 | Tied #1 (35) | 0 |
| HLEEpoch AI Benchmarking Hub | 56 | Clear #1 | 0 |
| LMArena AgentLMArena | 64 | Clear #1 | 0 |
| LMArena Document (style control)LMArena | 46 | Tied #1 (6) | 0 |
| LMArena DocumentLMArena | 46 | Tied #1 (4) | 0 |
| LMArena Image EditLMArena | 20 | Clear #1 | 0 |
| LMArena Image EditLMArena | 37 | Clear #1 | 0 |
| LMArena Image EditLMArena | 14 | Clear #1 | 0 |
| LMArena Image EditLMArena | 48 | Clear #1 | 0 |
| LMArena Image-to-VideoLMArena | 2 | Clear #1 | 0 |
| LMArena Search (style control)LMArena | 42 | Tied #1 (2) | 0 |
| LMArena SearchLMArena | 42 | Tied #1 (4) | 0 |
| LMArena Text (style control)LMArena | 395 | Tied #1 (5) | 0 |
| LMArena Text (style control)LMArena | 416 | Tied #1 (6) | 0 |
| LMArena Text (style control)LMArena | 419 | Tied #1 (3) | 0 |
| LMArena Text (style control)LMArena | 421 | Tied #1 (4) | 0 |
| LMArena Text (style control)LMArena | 421 | Tied #1 (2) | 0 |
| LMArena Text (style control)LMArena | 371 | Tied #1 (6) | 0 |
| LMArena Text (style control)LMArena | 301 | Tied #1 (6) | 0 |
| LMArena Text (style control)LMArena | 320 | Clear #1 | 0 |
| LMArena Text (style control)LMArena | 421 | Tied #1 (2) | 0 |
| LMArena Text (style control)LMArena | 419 | Tied #1 (9) | 0 |
| LMArena Text (style control)LMArena | 414 | Tied #1 (6) | 0 |
| LMArena Text (style control)LMArena | 419 | Tied #1 (2) | 0 |
| LMArena Text (style control)LMArena | 393 | Tied #1 (10) | 0 |
| LMArena Text (style control)LMArena | 419 | Tied #1 (8) | 0 |
| LMArena Text (style control)LMArena | 402 | Tied #1 (11) | 0 |
| LMArena Text (style control)LMArena | 391 | Tied #1 (20) | 0 |
| LMArena Text (style control)LMArena | 421 | Tied #1 (10) | 0 |
| LMArena Text (style control)LMArena | 420 | Tied #1 (3) | 0 |
| LMArena Text (style control)LMArena | 421 | Tied #1 (3) | 0 |
| LMArena Text (style control)LMArena | 288 | Tied #1 (6) | 0 |
| LMArena Text (style control)LMArena | 289 | Tied #1 (28) | 0 |
| LMArena Text (style control)LMArena | 399 | Tied #1 (2) | 0 |
| LMArena Text (style control)LMArena | 404 | Tied #1 (8) | 0 |
| LMArena Text (style control)LMArena | 419 | Tied #1 (2) | 0 |
| LMArena Text (style control)LMArena | 421 | Tied #1 (2) | 0 |
| LMArena Text (style control)LMArena | 421 | Tied #1 (2) | 0 |
| LMArena Text (style control)LMArena | 243 | Tied #1 (18) | 0 |
| LMArena Text (style control)LMArena | 385 | Tied #1 (3) | 0 |
| LMArena Text (style control)LMArena | 304 | Tied #1 (26) | 0 |
| LMArena Text-to-ImageLMArena | 85 | Tied #1 (2) | 0 |
| LMArena Text-to-ImageLMArena | 28 | Tied #1 (2) | 0 |
| LMArena Text-to-ImageLMArena | 85 | Tied #1 (2) | 0 |
| LMArena Text-to-ImageLMArena | 85 | Tied #1 (2) | 0 |
| LMArena Text-to-ImageLMArena | 8 | Tied #1 (3) | 0 |
| LMArena Text-to-ImageLMArena | 85 | Clear #1 | 0 |
| LMArena Text-to-ImageLMArena | 8 | Clear #1 | 0 |
| LMArena Text-to-ImageLMArena | 85 | Clear #1 | 0 |
| LMArena Text-to-ImageLMArena | 85 | Clear #1 | 0 |
| LMArena Text-to-ImageLMArena | 85 | Tied #1 (4) | 0 |
| LMArena Text-to-ImageLMArena | 8 | Tied #1 (2) | 0 |
| LMArena Text-to-ImageLMArena | 85 | Clear #1 | 0 |
| LMArena Text-to-VideoLMArena | 2 | Clear #1 | 0 |
| LMArena TextLMArena | 395 | Tied #1 (3) | 0 |
| LMArena TextLMArena | 416 | Tied #1 (2) | 0 |
| LMArena TextLMArena | 419 | Tied #1 (3) | 0 |
| LMArena TextLMArena | 421 | Tied #1 (3) | 0 |
| LMArena TextLMArena | 421 | Tied #1 (2) | 0 |
| LMArena TextLMArena | 371 | Tied #1 (6) | 0 |
| LMArena TextLMArena | 301 | Tied #1 (2) | 0 |
| LMArena TextLMArena | 320 | Clear #1 | 0 |
| LMArena TextLMArena | 421 | Tied #1 (2) | 0 |
| LMArena TextLMArena | 419 | Tied #1 (5) | 0 |
| LMArena TextLMArena | 414 | Tied #1 (4) | 0 |
| LMArena TextLMArena | 419 | Tied #1 (4) | 0 |
| LMArena TextLMArena | 393 | Tied #1 (13) | 0 |
| LMArena TextLMArena | 419 | Tied #1 (4) | 0 |
| LMArena TextLMArena | 402 | Tied #1 (8) | 0 |
| LMArena TextLMArena | 391 | Tied #1 (8) | 0 |
| LMArena TextLMArena | 421 | Tied #1 (2) | 0 |
| LMArena TextLMArena | 420 | Tied #1 (3) | 0 |
| LMArena TextLMArena | 421 | Tied #1 (4) | 0 |
| LMArena TextLMArena | 288 | Tied #1 (5) | 0 |
| LMArena TextLMArena | 289 | Tied #1 (13) | 0 |
| LMArena TextLMArena | 399 | Tied #1 (2) | 0 |
| LMArena TextLMArena | 404 | Tied #1 (7) | 0 |
| LMArena TextLMArena | 419 | Tied #1 (2) | 0 |
| LMArena TextLMArena | 421 | Tied #1 (2) | 0 |
| LMArena TextLMArena | 421 | Tied #1 (2) | 0 |
| LMArena TextLMArena | 243 | Tied #1 (15) | 0 |
| LMArena TextLMArena | 385 | Tied #1 (2) | 0 |
| LMArena TextLMArena | 304 | Tied #1 (9) | 0 |
| LMArena Vision (style control)LMArena | 34 | Tied #1 (8) | 0 |
| LMArena Vision (style control)LMArena | 135 | Tied #1 (18) | 0 |
| LMArena Vision (style control)LMArena | 34 | Clear #1 | 0 |
| LMArena Vision (style control)LMArena | 106 | Tied #1 (8) | 0 |
| LMArena Vision (style control)LMArena | 123 | Tied #1 (3) | 0 |
| LMArena Vision (style control)LMArena | 167 | Tied #1 (11) | 0 |
| LMArena Vision (style control)LMArena | 52 | Tied #1 (10) | 0 |
| LMArena Vision (style control)LMArena | 118 | Tied #1 (10) | 0 |
| LMArena Vision (style control)LMArena | 102 | Tied #1 (8) | 0 |
| LMArena Vision (style control)LMArena | 123 | Tied #1 (2) | 0 |
| LMArena Vision (style control)LMArena | 167 | Tied #1 (3) | 0 |
| LMArena VisionLMArena | 34 | Tied #1 (7) | 0 |
| LMArena VisionLMArena | 135 | Tied #1 (8) | 0 |
| LMArena VisionLMArena | 34 | Clear #1 | 0 |
| LMArena VisionLMArena | 106 | Tied #1 (12) | 0 |
| LMArena VisionLMArena | 123 | Tied #1 (9) | 0 |
| LMArena VisionLMArena | 167 | Tied #1 (8) | 0 |
| LMArena VisionLMArena | 52 | Tied #1 (6) | 0 |
| LMArena VisionLMArena | 118 | Tied #1 (5) | 0 |
| LMArena VisionLMArena | 102 | Tied #1 (8) | 0 |
| LMArena VisionLMArena | 123 | Tied #1 (5) | 0 |
| LMArena VisionLMArena | 167 | Tied #1 (6) | 0 |
| LMArena WebDevLMArena | 64 | Tied #1 (2) | 0 |
| LMArena WebDevLMArena | 156 | Tied #1 (2) | 0 |
| LMArena WebDevLMArena | 156 | Tied #1 (2) | 0 |
| LMArena WebDevLMArena | 155 | Tied #1 (4) | 0 |
| LMArena WebDevLMArena | 140 | Tied #1 (2) | 0 |
| METR time horizonMETR time horizons | 3 | Clear #1 | 0 |
| METR time horizonMETR time horizons | 23 | Tied #1 (23) | 0 |
| OSWorldEpoch AI Benchmarking Hub | 38 | Clear #1 | 0 |
| SWE-Bench verifiedEpoch AI Benchmarking Hub | 35 | Tied #1 (5) | 0 |
| Terminal BenchEpoch AI Benchmarking Hub | 202 | Tied #1 (8) | 0 |
| Code MigrationVals AI benchmarks | 38 | Tied #1 (5) | 0 |
| EMBVals AI benchmarks | 38 | Tied #1 (16) | 0 |
| Finance AgentVals AI benchmarks | 38 | Tied #1 (15) | 0 |
| Vals IndexVals AI benchmarks | 38 | Tied #1 (5) | 0 |
| Legal Agent Benchmark (HLAB-Vals)Vals AI benchmarks | 38 | Tied #1 (8) | 0 |
| Legal ResearchVals AI benchmarks | 38 | Tied #1 (12) | 0 |
| Tax AgentVals AI benchmarks | 38 | Tied #1 (8) | 0 |
| Terminal-Bench 4Vals AI benchmarks | 38 | Tied #1 (7) | 0 |
| Vibe Code BenchVals AI benchmarks | 38 | Tied #1 (7) | 0 |
| Vending-Bench 2Epoch AI Benchmarking Hub | 72 | Clear #1 | 0 |
| Model | Grade | Independent sources | Vendor sources | Recency |
|---|---|---|---|---|
| DeepSeek V4.1 Flash 1017 | Grade A. Evidence grade A of A to D: strongest. | 4 | 1 | 2d (recent) |
| Fable 5.2 | Grade A. Evidence grade A of A to D: strongest. | 4 | 1 | 2d (recent) |
| Gemini 3.8 Flash | Grade A. Evidence grade A of A to D: strongest. | 4 | 1 | 2d (recent) |
| Gemini 4 Argon | Grade A. Evidence grade A of A to D: strongest. | 4 | 2 | 2d (recent) |
| Gemini 4 Carbon | Grade A. Evidence grade A of A to D: strongest. | 4 | 1 | 2d (recent) |
| GLM-5.3 | Grade A. Evidence grade A of A to D: strongest. | 4 | 2 | 2d (recent) |
| GLM-5.3 Flash | Grade A. Evidence grade A of A to D: strongest. | 4 | 2 | 2d (recent) |
| GPT-6 Astra | Grade A. Evidence grade A of A to D: strongest. | 4 | 1 | 2d (recent) |
| GPT-6 Luna | Grade A. Evidence grade A of A to D: strongest. | 4 | 1 | 2d (recent) |
| GPT-6.1 Sol | Grade A. Evidence grade A of A to D: strongest. | 4 | 1 | 2d (recent) |
| GPT-6.1 Terra | Grade A. Evidence grade A of A to D: strongest. | 4 | 0 | 2d (recent) |
| Grok 4.7 | Grade A. Evidence grade A of A to D: strongest. | 4 | 2 | 2d (recent) |
| Haiku 5.5 | Grade A. Evidence grade A of A to D: strongest. | 4 | 2 | 2d (recent) |
| Kimi K3 | Grade A. Evidence grade A of A to D: strongest. | 4 | 1 | 2d (recent) |
| Kimi K3.1 Pro | Grade A. Evidence grade A of A to D: strongest. | 4 | 0 | 2d (recent) |
| Muse Wonder 1.0 | Grade A. Evidence grade A of A to D: strongest. | 4 | 1 | 2d (recent) |
| Opus 5.5 | Grade A. Evidence grade A of A to D: strongest. | 4 | 2 | 2d (recent) |
| Sonnet 5.5 | Grade A. Evidence grade A of A to D: strongest. | 4 | 2 | 2d (recent) |
| Beam 1.0 | Grade A. Evidence grade A of A to D: strongest. | 3 | 0 | 2d (recent) |
| Claude 3 Opus | Grade A. Evidence grade A of A to D: strongest. | 3 | 0 | 6d (recent) |
| Claude 3.5 Sonnet (June '24) | Grade A. Evidence grade A of A to D: strongest. | 3 | 0 | 6d (recent) |
| Claude 3.5 Sonnet (Oct '24) | Grade A. Evidence grade A of A to D: strongest. | 3 | 0 | 6d (recent) |
| Claude 3.7 Sonnet | Grade A. Evidence grade A of A to D: strongest. | 3 | 0 | 6d (recent) |
| DeepSeek V4 Pro 0813 | Grade A. Evidence grade A of A to D: strongest. | 3 | 1 | 2d (recent) |
| DeepSeek V4-Pro | Grade A. Evidence grade A of A to D: strongest. | 3 | 1 | 2d (recent) |
| DeepSeek V4.1 Flash | Grade A. Evidence grade A of A to D: strongest. | 3 | 1 | 2d (recent) |
| Fable 5 | Grade A. Evidence grade A of A to D: strongest. | 3 | 1 | 2d (recent) |
| Fable 5.1 | Grade A. Evidence grade A of A to D: strongest. | 3 | 2 | 2d (recent) |
| Gemini 3 Pro Preview | Grade A. Evidence grade A of A to D: strongest. | 3 | 1 | 2d (recent) |
| Gemini 3.1 Pro Preview | Grade A. Evidence grade A of A to D: strongest. | 3 | 0 | 2d (recent) |
| Gemini 3.7 Flash | Grade A. Evidence grade A of A to D: strongest. | 3 | 1 | 2d (recent) |
| Gemini 4 Hydrogen | Grade A. Evidence grade A of A to D: strongest. | 3 | 1 | 2d (recent) |
| Gemma 4 31B | Grade A. Evidence grade A of A to D: strongest. | 3 | 0 | 2d (recent) |
| GPT-4 (0314) | Grade A. Evidence grade A of A to D: strongest. | 3 | 0 | 6d (recent) |
| GPT-4 Turbo | Grade A. Evidence grade A of A to D: strongest. | 3 | 0 | 6d (recent) |
Not enough data: 1 vendor-run board (need 2).
| Lab | Paired models | Median vendor pct | Median independent pct | Signal |
|---|---|---|---|---|
| alibaba | 1 | 4.4 | 73.7 | Insufficient sample |
| anthropic | 15 | 72.8 | 90.9 | no-clear-difference |
| deepseek | 5 | 28.9 | 70.3 | lower-on-vendor-boards |
| 8 | 61.1 | 86.1 | lower-on-vendor-boards | |
| meta | 2 | 47.5 | 81.2 | Insufficient sample |
| minimax | 1 | 11.1 | 56.0 | Insufficient sample |
| mistral | 1 | 40.0 | 64.1 | Insufficient sample |
| moonshot-ai | 2 | 48.9 | 54.4 | Insufficient sample |
| openai | 14 | 51.7 | 67.0 | no-clear-difference |
| thinking-machines | 1 | 8.9 | 5.4 | Insufficient sample |
| xai | 4 | 54.0 | 66.5 | lower-on-vendor-boards |
| zai | 3 | 24.2 | 68.9 | lower-on-vendor-boards |
Method rules
Known limits
Exclusions