Acropolis

Benchmark directory

Benchmarks

Filter benchmarks by capability domain or evaluation independence, inspect the current leader with uncertainty bounds, or open any board to explore full results.

Domains
Showing 52 of 52 benchmarks
  • LMArena Text

    LMArena · 29 slices · 11,177 rows

    Independent. Run by an operator independent of the model vendors.

    Arena human-vote leaderboard: rating with confidence interval, vote count and rank. One row per model and category.

    General text
    Current #1 by arena ratingTied #1 (2)

    Gemini 4 Carbon (Google)

    1,538[1,529 to 1,548]

    Fetched 17 Nov 2026Open board
  • LMArena Text (style control)

    LMArena · 29 slices · 11,177 rows

    Independent. Run by an operator independent of the model vendors.

    Arena human-vote leaderboard: rating with confidence interval, vote count and rank. One row per model and category.

    General text
    Current #1 by arena ratingTied #1 (2)

    Gemini 4 Carbon (Google)

    1,526[1,517 to 1,536]

    Fetched 17 Nov 2026Open board
  • LMArena Vision

    LMArena · 11 slices · 1,161 rows

    Independent. Run by an operator independent of the model vendors.

    Arena human-vote leaderboard: rating with confidence interval, vote count and rank. One row per model and category.

    Vision
    Current #1 by arena ratingTied #1 (6)

    Fable 5 (Anthropic)

    1,326[1,319 to 1,332]

    Fetched 17 Nov 2026Open board
  • LMArena Vision (style control)

    LMArena · 11 slices · 1,161 rows

    Independent. Run by an operator independent of the model vendors.

    Arena human-vote leaderboard: rating with confidence interval, vote count and rank. One row per model and category.

    Vision
    Current #1 by arena ratingTied #1 (3)

    Fable 5 (Anthropic)

    1,309[1,303 to 1,316]

    Fetched 17 Nov 2026Open board
  • LMArena Text-to-Image

    LMArena · 12 slices · 732 rows

    Independent. Run by an operator independent of the model vendors.

    Arena human-vote leaderboard: rating with confidence interval, vote count and rank. One row per model and category.

    Image generation
    Current #1 by arena rating

    GPT Image 2.5 Sunburst (OpenAI)

    1,429[1,424 to 1,434]

    Fetched 17 Nov 2026Open board
  • LMArena WebDev

    LMArena · 5 slices · 671 rows

    Independent. Run by an operator independent of the model vendors.

    Arena human-vote leaderboard: rating with confidence interval, vote count and rank. One row per model and category.

    Coding
    Current #1 by arena ratingTied #1 (2)

    Fable 5.2 (Anthropic)

    1,826[1,807 to 1,845]

    Fetched 17 Nov 2026Open board
  • GPQA diamond

    Epoch AI Benchmarking Hub · 338 rows

    Independent. Run by an operator independent of the model vendors.

    GPQA Diamond, run by Epoch AI.

    Science
    Current #1 by accuracy (best score across scorers)Tied #1 (35)

    Gemini 4 Carbon (Google)

    96.5%[94.2% to 98.9%]

    Fetched 17 Nov 2026Open board
  • ARC-AGI-2

    Epoch AI Benchmarking Hub · 299 rows

    Independent. Run by an operator independent of the model vendors.

    ARC-AGI-2 results that Epoch AI republishes from the original runner.

    General textScience
    Current #1 by score

    GPT-6 Astra (OpenAI)

    95%

    Fetched 17 Nov 2026Open board
  • Terminal Bench

    Epoch AI Benchmarking Hub · 202 rows

    Independent. Run by an operator independent of the model vendors.

    Terminal-Bench results from its public leaderboard that Epoch AI republishes. Each row is a model run in an agent harness.

    CodingAgentic
    Current #1 by accuracyTied #1 (8)

    GPT-5.5 (OpenAI)

    84.7%[80.6% to 88.8%]

    Fetched 17 Nov 2026Open board
  • Bug Hunt Bench

    Bug Hunt Bench · 145 rows

    Independent. Run by an operator independent of the model vendors.

    105 real bugs planted in two production repos. Coding agents find and fix them; a blind judge grades the diff against a withheld answer key.

    CodingAgentic
    Current #1 by bugs fixed

    Fable 5.2 (Anthropic)

    58

    Fetched 14 Nov 2026Open board
  • LMArena Image Edit

    LMArena · 4 slices · 119 rows

    Independent. Run by an operator independent of the model vendors.

    Arena human-vote leaderboard: rating with confidence interval, vote count and rank. One row per model and category.

    Image generationVision
    Current #1 by arena rating

    GPT Image 2.5 Sunburst (OpenAI)

    1,519[1,514 to 1,523]

    Fetched 17 Nov 2026Open board
  • FrontierMath-2025-02-28-Private

    Epoch AI Benchmarking Hub · 101 rows

    Independent. Run by an operator independent of the model vendors.

    FrontierMath, 2025-02-28 private set, run by Epoch AI. Superseded in benchmark_metadata by FrontierMath-Tiers-1-3-v2-Private.

    Math
    Current #1 by accuracy (best score across scorers)Tied #1 (6)

    GPT-5.5 Pro Pre-release (OpenAI)

    52.4%[46.7% to 58.1%]

    Fetched 17 Nov 2026Open board
  • CursorBench

    CursorBench · 91 rows

    Vendor-run. Run by the vendor of at least one model on the list.

    Agents are evaluated on ambiguous, multi-file tasks from real Cursor sessions. Run by Cursor (a vendor that also sells its own model, Composer).

    CodingAgentic
    Current #1 by score

    Gemini 4 Carbon (Google)

    60.4%

    Fetched 17 Nov 2026Open board
  • Vending-Bench 2

    Epoch AI Benchmarking Hub · 72 rows

    Independent. Run by an operator independent of the model vendors.

    Vending-Bench 2 results that Epoch AI republishes. Scores go from about -31 to 16484, so they are not 0 to 1.

    AgenticFinance
    Current #1 by score

    Gemini 4 Carbon (Google)

    $16,484.21

    Fetched 17 Nov 2026Open board
  • LMArena Agent

    LMArena · 64 rows

    Independent. Run by an operator independent of the model vendors.

    Arena agent leaderboard: score with confidence interval, cost and output tokens per task, observation and session counts. One row per model and category.

    AgenticCoding
    Current #1 by agent score

    Fable 5.2 (Anthropic)

    0.3[0.2 to 0.3]

    Fetched 17 Nov 2026Open board
  • HLE

    Epoch AI Benchmarking Hub · 56 rows

    Independent. Run by an operator independent of the model vendors.

    Humanity's Last Exam results that Epoch AI republishes from the original runner.

    ScienceMathGeneral text
    Current #1 by accuracy

    GPT-6 Astra (OpenAI)

    54.8%[51% to 58.6%]

    Fetched 17 Nov 2026Open board
  • FrontierCode

    Epoch AI Benchmarking Hub · 51 rows

    Independent. Run by an operator independent of the model vendors.

    Merge-quality coding score. Run by Cognition (a vendor); Epoch republishes the results. Aggregation per row, for example Mean@5.

    CodingAgentic
    Current #1 by main score

    Opus 5.5 (Anthropic)

    54.6%

    Fetched 17 Nov 2026Open board
  • LMArena Document

    LMArena · 46 rows

    Independent. Run by an operator independent of the model vendors.

    Arena human-vote leaderboard: rating with confidence interval, vote count and rank. One row per model and category.

    General text
    Current #1 by arena ratingTied #1 (4)

    Fable 5.1 (Anthropic)

    1,515[1,500 to 1,530]

    Fetched 17 Nov 2026Open board
  • Independent. Run by an operator independent of the model vendors.

    Arena human-vote leaderboard: rating with confidence interval, vote count and rank. One row per model and category.

    General text
    Current #1 by arena ratingTied #1 (6)

    Fable 5 (Anthropic)

    1,500[1,493 to 1,507]

    Fetched 17 Nov 2026Open board
  • LMArena Search

    LMArena · 42 rows

    Independent. Run by an operator independent of the model vendors.

    Arena human-vote leaderboard: rating with confidence interval, vote count and rank. One row per model and category.

    Search
    Current #1 by arena ratingTied #1 (4)

    Qwen3.8-Max (Alibaba)

    1,260[1,246 to 1,273]

    Fetched 17 Nov 2026Open board
  • Independent. Run by an operator independent of the model vendors.

    Arena human-vote leaderboard: rating with confidence interval, vote count and rank. One row per model and category.

    Search
    Current #1 by arena ratingTied #1 (2)

    GPT-5.6 Sol (OpenAI)

    1,255[1,249 to 1,262]

    Fetched 17 Nov 2026Open board
  • Code Migration

    Vals AI benchmarks · 38 rows

    Independent. Run by an operator independent of the model vendors.

    Code Migration, a component of the Vals Index.

    CodingAgentic
    Current #1 by accuracyTied #1 (5)

    Sonnet 5.5 (Anthropic)

    87.4%[82.7% to 92.1%]

    Fetched 17 Nov 2026Open board
  • EMB

    Vals AI benchmarks · 38 rows

    Independent. Run by an operator independent of the model vendors.

    EMB, a component of the Vals Index.

    FinanceAgentic
    Current #1 by accuracyTied #1 (16)

    Fable 5.2 (Anthropic)

    77.4%[73.1% to 81.7%]

    Fetched 17 Nov 2026Open board
  • Finance Agent

    Vals AI benchmarks · 38 rows

    Independent. Run by an operator independent of the model vendors.

    Finance Agent, a component of the Vals Index.

    FinanceAgentic
    Current #1 by accuracyTied #1 (15)

    Fable 5.2 (Anthropic)

    69.5%[65.6% to 73.3%]

    Fetched 17 Nov 2026Open board
  • Legal Agent Benchmark (HLAB-Vals)

    Vals AI benchmarks · 38 rows

    Independent. Run by an operator independent of the model vendors.

    Legal Agent Benchmark (HLAB-Vals), a component of the Vals Index.

    LegalAgentic
    Current #1 by accuracyTied #1 (8)

    Gemini 4 Argon (Google)

    27.1%[19.8% to 34.4%]

    Fetched 17 Nov 2026Open board
  • Legal Research

    Vals AI benchmarks · 38 rows

    Independent. Run by an operator independent of the model vendors.

    Legal Research, a component of the Vals Index.

    LegalAgentic
    Current #1 by accuracyTied #1 (12)

    Gemini 4 Carbon (Google)

    55.8%[49% to 62.5%]

    Fetched 17 Nov 2026Open board
  • OSWorld

    Epoch AI Benchmarking Hub · 38 rows

    Independent. Run by an operator independent of the model vendors.

    OSWorld results from the OSWorld site and vendor posts that Epoch AI republishes. The Agent column carries the step budget, for example `(100 steps)`. Scores are already 0 to 100.

    AgenticVision
    Current #1 by score

    Sonnet 4.6 (Anthropic)

    72.1%

    Fetched 17 Nov 2026Open board
  • Tax Agent

    Vals AI benchmarks · 38 rows

    Independent. Run by an operator independent of the model vendors.

    Tax Agent, a component of the Vals Index.

    FinanceAgentic
    Current #1 by accuracyTied #1 (8)

    Gemini 4 Hydrogen (Google)

    52.3%[45.2% to 59.4%]

    Fetched 17 Nov 2026Open board
  • Terminal-Bench 4

    Vals AI benchmarks · 38 rows

    Independent. Run by an operator independent of the model vendors.

    Terminal-Bench 4, a component of the Vals Index.

    CodingAgentic
    Current #1 by accuracyTied #1 (7)

    Fable 5.2 (Anthropic)

    65.2%[60.6% to 69.7%]

    Fetched 17 Nov 2026Open board
  • Vals Index

    Vals AI benchmarks · 38 rows

    Independent. Run by an operator independent of the model vendors.

    A single measure of AI's potential economic impact: agentic model performance across finance, coding, legal, and tax tasks, weighted by each sector's share of U.S. GDP.

    AgenticFinanceCodingLegal
    Current #1 by accuracyTied #1 (5)

    Fable 5.2 (Anthropic)

    71.3%[69.4% to 73.2%]

    Fetched 17 Nov 2026Open board
  • Vibe Code Bench

    Vals AI benchmarks · 38 rows

    Independent. Run by an operator independent of the model vendors.

    Vibe Code Bench, a component of the Vals Index.

    CodingAgentic
    Current #1 by accuracyTied #1 (7)

    Fable 5.2 (Anthropic)

    93.8%[91.5% to 96%]

    Fetched 17 Nov 2026Open board
  • SWE-Bench verified

    Epoch AI Benchmarking Hub · 35 rows

    Independent. Run by an operator independent of the model vendors.

    SWE-bench Verified, run by Epoch AI.

    CodingAgentic
    Current #1 by accuracy (best score across scorers)Tied #1 (5)

    Opus 4.7 (Anthropic)

    83.5%[80.2% to 86.8%]

    Fetched 17 Nov 2026Open board
  • METR time horizon

    METR time horizons · 2 slices · 26 rows

    Independent. Run by an operator independent of the model vendors.

    Length of tasks, measured in human working time, that a model completes with 50 percent or 80 percent success. Run on METR's long-task suite.

    AgenticCoding
    Current #1 by 50% time horizonTied #1 (23)

    Claude Mythos Preview (early) (Anthropic)

    1,044.8 min[508.9 min to 3,304.3 min]

    Fetched 15 Nov 2026Open board
  • GDPval

    Epoch AI Benchmarking Hub · 11 rows

    Independent. Run by an operator independent of the model vendors.

    GDPval results that Epoch AI republishes. Despite the column label (%), the values are 0 to 1 fractions.

    FinanceGeneral text
    Current #1 by win rate

    GPT-5.2 (OpenAI)

    49.7%

    Fetched 17 Nov 2026Open board
  • LMArena Image-to-Video

    LMArena · 2 rows

    Independent. Run by an operator independent of the model vendors.

    Arena human-vote leaderboard: rating with confidence interval, vote count and rank. One row per model and category.

    Video generationVision
    Current #1 by arena rating

    Wan 2.5 Preview (Alibaba)

    1,322[1,312 to 1,332]

    Fetched 17 Nov 2026Open board
  • LMArena Text-to-Video

    LMArena · 2 rows

    Independent. Run by an operator independent of the model vendors.

    Arena human-vote leaderboard: rating with confidence interval, vote count and rank. One row per model and category.

    Video generation
    Current #1 by arena rating

    Wan 2.6 Image (Alibaba)

    1,324[1,316 to 1,331]

    Fetched 17 Nov 2026Open board
  • Artificial Analysis Image editing arena

    Artificial Analysis · 0 rows

    Independent. Run by an operator independent of the model vendors.Link-only in public build

    Elo from human preference votes, with the 95 percent interval the source reports.

    Image generationVision
    Redistribution restricted by source license. View scores directly at Artificial Analysis.
    Fetched 15 Nov 2026Open board
  • Artificial Analysis Image to video arena

    Artificial Analysis · 0 rows

    Independent. Run by an operator independent of the model vendors.Link-only in public build

    Elo from human preference votes, with the 95 percent interval the source reports.

    Video generationVision
    Redistribution restricted by source license. View scores directly at Artificial Analysis.
    Fetched 15 Nov 2026Open board
  • Artificial Analysis Intelligence Index

    Artificial Analysis · 0 rows

    Independent. Run by an operator independent of the model vendors.Link-only in public build

    Composite index of language model quality, with the cost to run it.

    General text
    Redistribution restricted by source license. View scores directly at Artificial Analysis.
    Fetched 15 Nov 2026Open board
  • Artificial Analysis Text to image arena

    Artificial Analysis · 0 rows

    Independent. Run by an operator independent of the model vendors.Link-only in public build

    Elo from human preference votes, with the 95 percent interval the source reports.

    Image generation
    Redistribution restricted by source license. View scores directly at Artificial Analysis.
    Fetched 15 Nov 2026Open board
  • Artificial Analysis Text to speech arena

    Artificial Analysis · 0 rows

    Independent. Run by an operator independent of the model vendors.Link-only in public build

    Elo from human preference votes, with the 95 percent interval the source reports.

    Speech
    Redistribution restricted by source license. View scores directly at Artificial Analysis.
    Fetched 15 Nov 2026Open board
  • Artificial Analysis Text to video arena

    Artificial Analysis · 0 rows

    Independent. Run by an operator independent of the model vendors.Link-only in public build

    Elo from human preference votes, with the 95 percent interval the source reports.

    Video generation
    Redistribution restricted by source license. View scores directly at Artificial Analysis.
    Fetched 15 Nov 2026Open board
  • Artificial Analysis performance

    Artificial Analysis · 0 rows

    Independent. Run by an operator independent of the model vendors.Link-only in public build

    Output speed and latency. Median values come from the `performance` block, mean values from `performance_v3`.

    Speed and latency
    Redistribution restricted by source license. View scores directly at Artificial Analysis.
    Fetched 15 Nov 2026Open board
  • Artificial Analysis token prices

    Artificial Analysis · 0 rows

    Independent. Run by an operator independent of the model vendors.Link-only in public build

    List prices per million tokens.

    Speed and latency
    Redistribution restricted by source license. View scores directly at Artificial Analysis.
    Fetched 15 Nov 2026Open board
Catalog benchmarks without results in this build (1)
  • LMArena Video Edit