Benchmark Almanac

Compare models

Pick up to four models to see how they stack up on every benchmark they've all been scored on.

Scores from

Who comes out ahead?

Claude Fable 5.2 leads on 26 of the 62 benchmarks all four share. We use independent runs where they exist and fill the gaps with labs’ own numbers. 5 benchmarks are a tie for first.

  1. GPT-6.2 AstraBest on 13, tied for best on 4
  2. Claude Fable 5.2Best on 26, tied for best on 4
  3. Gemini 4 CarbonBest on 17, tied for best on 2
  4. Muse WonderBest on 1

Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards

Cited by
Labs
Area
AutomationBench52.451.553.939.1 (the lab’s own number)204 of 414.8
FrontierCode 2.044.250.242.625.382 of 424.9
LiveBench84.984.983.370.780 of 414.2
LMArena (Elo)Elo1,4681,5031,5271,51170 of 459
Agents' Last Exam v1.147.546.945.938.552 of 49.0
Agents' Last Exam v1.1 (Score)78.372.877.468.452 of 410.0
Artificial Analysis Intelligence Index v5index62.164.660.043.632 of 421.0
HLE-Diamond72.377.963.640.630 of 437.3
Legal Research Bench42.846.155.850.530 of 413.0
Vals Index v2.267.771.370.564.921 of 46.4
Harvey LAB v1.1 (HLAB-Vals-1.1)16.35.0017.523.820 of 418.8
Tax Agent Bench21.832.151.836.320 of 429.9
Tax Agent Bench (Weighted pass)57.967.076.471.220 of 418.6
Vibe Code Bench89.993.889.183.120 of 410.7
APEX-Accounting69.870.470.767.110 of 43.6
APEX-Accounting (Pass@1)19.522.522.521.310 of 43.0
CWE-bench 266.875.067.366.110 of 48.9
DesignArena Game DevElo1,4501,3921,3291,40910 of 4121
EEBench74.677.383.962.710 of 421.2
Excel Modeling Benchmark v1.176.077.476.572.510 of 44.9
FrontierMath Tier 4 v2100.0100.095.170.710 of 429.3
FrontierMath Tiers 1-3 v292.695.190.582.110 of 413.0
Humanity's Sixth Sense63.259.571.143.710 of 427.4
Terminal-Bench 4.162.165.864.547.010 of 418.8
Vals Finance Agent v2.160.869.567.265.510 of 48.7
Vals Finance Index57.363.667.061.210 of 49.7
Vals Legal Index69.769.275.975.310 of 46.8
APEX-Agents 1.183.987.690.477.000 of 413.4
Argo-Benchscore62.970.262.337.600 of 432.6
BullshitBench V255.067.372.057.600 of 417.0
BuseyBench SVG Testscore7.407.207.906.6000 of 41.30
CAIS Risk Indexscore, lower is better40.58.4033.432.400 of 432.1
CAIS Text Capabilities Indexscore54.766.359.047.600 of 418.7
ContractorOS Quantity Takeoff53.859.260.031.500 of 428.5
DataAnalyticsBenchscore41.043.042.041.000 of 42.00
DAXBench v293.687.892.882.200 of 411.4
DAYJOB: Accounting42.045.225.230.800 of 420.0
DAYJOB: Law41.426.513.820.600 of 427.6
DeepFinance61.462.056.644.200 of 417.8
DesignArena PDFElo1,2821,3461,3211,22500 of 4121
DualEntry Accounting AI Benchmark78.281.288.183.200 of 49.9
EBR-bench92.490.587.631.400 of 460.9
Extended NYT Connections98.297.399.393.400 of 45.9
Fiction Bench (Story Hacker)score73.662.164.464.000 of 411.5
Furniture Assembly Benchmark89.290.077.554.200 of 435.8
GradeBench56.357.356.648.100 of 49.2
KradleBenchscore78.337.822.027.600 of 456.3
Lech Mazur Creative Writingscore4.303.561.621.6100 of 42.69
LLM Debate Benchmarkrating1,6731,6881,6761,63700 of 451
Maze Bench17.022.016.00.000 of 422.0
Mystery Game Puzzles95.090.073.051.000 of 444.0
OTIS Mock AIME 2024-2025100.099.7100.099.400 of 40.6
OWE-Bench67.254.632.812.700 of 454.5
Parallel Search Capability Leaderboardscore84.586.985.171.300 of 415.6
Ramen Benchrating443200 of 42
Ramp Accounting Bench50.452.256.548.700 of 47.8
Ramp SWE-Bench v276.473.474.058.600 of 417.8
SWE Atlas v1.160.580.571.564.800 of 420.1
TaxCalcBench v276.086.075.084.000 of 411.0
The Aggregate Personal Style Imitation75.573.067.064.900 of 410.6
Toloka Arena Enterprise Tool Use (November 2026)72.077.074.065.700 of 411.3
Tuesday Work Indexindex70.371.265.663.800 of 47.4

A lab’s own number. Everything else is an independent run.Scores are percentages unless the row says otherwise. Bold is the best in each row. Click a row for its bars.

Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards

Gap is how far apart the best and worst picked models are on a test. A small gap means they're about even. A big one means the test tells them apart.

It's in points for percentages (74.3 against 68.1 is 6.2) and in the test's own unit otherwise, such as Elo. To sort mixed units together, we count those as a share of the test's full range.

Labs is how many labs report this benchmark in their launch posts and system cards, across the whole almanac. A high count means a widely used test.

Cited by is how many of the picked models' own launch posts and system cards report this benchmark. At 0, every score in the row is an independent run.