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 21 of the 45 benchmarks all four share. We use independent runs where they exist and fill the gaps with labs’ own numbers. One benchmark is a tie for first.
- Claude Fable 5.2Best on 21
Gemini 4 CarbonBest on 13, tied for best on 1- GPT-6 AstraBest on 6, tied for best on 1
Muse WonderBest on 4
Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards
No shared benchmark matches these filters.
| AutomationBench | 51.5 | 53.9 | 41.4 | 39.1 (the lab’s own number) | 20 | 4 of 4 | 14.8 |
| Terminal-Bench 4.0 | 67.6 (the lab’s own number) | 65.2 (the lab’s own number) | 58.2 | 47.0 | 16 | 4 of 4 | 20.6 |
| LiveBench | 84.9 | 83.5 | 84.5 | 70.7 | 8 | 0 of 4 | 14.2 |
| FrontierCode 2.0 | 50.2 | 42.6 | 44.0 | 25.3 | 6 | 1 of 4 | 24.9 |
| Code ArenaElo | 1,836 | 1,768 | 1,764 | 1,649 | 4 | 0 of 4 | 187 |
| PostTrainBench 1.2 | 44.6 | 41.0 | 41.9 | 24.2 | 3 | 0 of 4 | 20.4 |
| Blueprint-Bench 2 | 56.6 | 58.6 | 49.7 | 31.7 | 2 | 1 of 4 | 26.9 |
| Vals Index v2.2 | 71.3 | 70.5 | 65.0 | 64.9 | 2 | 1 of 4 | 6.4 |
| HLE-Diamond | 77.9 | 63.6 | 65.8 | 40.6 | 2 | 0 of 4 | 37.3 |
| Legal Research Bench | 46.1 | 55.8 | 39.4 | 50.5 | 2 | 0 of 4 | 16.4 |
| Vibe Code Bench | 93.8 | 89.1 | 89.6 | 83.1 | 2 | 0 of 4 | 10.7 |
| Agent Arena | 30.4 | 17.0 | 14.5 | 1.77 | 1 | 0 of 4 | 28.6 |
| APEX-Accounting | 70.4 | 70.7 | 64.7 | 67.1 | 1 | 0 of 4 | 6.0 |
| DesignArena 3D DesignElo | 1,516 | 1,419 | 1,525 | 1,434 | 1 | 0 of 4 | 106 |
| DesignArena Game DevElo | 1,399 | 1,322 | 1,470 | 1,416 | 1 | 0 of 4 | 148 |
| DesignArena WebsiteElo | 1,352 | 1,293 | 1,341 | 1,380 | 1 | 0 of 4 | 87 |
| EEBench | 77.3 | 83.9 | 69.3 | 62.7 | 1 | 0 of 4 | 21.2 |
| Excel Modeling Benchmark v1.1 | 77.4 | 76.5 | 72.7 | 72.5 | 1 | 0 of 4 | 4.9 |
| FrontierMath Tier 4 v2 | 100.0 | 95.1 | 97.6 | 70.7 | 1 | 0 of 4 | 29.3 |
| FrontierMath Tiers 1-3 v2 | 95.1 | 90.5 | 93.7 | 82.1 | 1 | 0 of 4 | 13.0 |
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 5.00 | 17.5 | 7.08 | 23.8 | 1 | 0 of 4 | 18.8 |
| Humanity's Sixth Sense | 59.5 | 71.1 | 53.6 | 43.7 | 1 | 0 of 4 | 27.4 |
| PRBench | 49.0 | 53.7 | 48.4 | 58.4 | 1 | 0 of 4 | 10.0 |
| Tax Agent Bench | 32.1 | 51.8 | 20.7 | 36.3 | 1 | 0 of 4 | 31.1 |
| Vals Finance Agent v2.1 | 69.5 | 67.2 | 60.4 | 65.5 | 1 | 0 of 4 | 9.0 |
| Vals Finance Index | 63.6 | 67.0 | 55.6 | 61.2 | 1 | 0 of 4 | 11.4 |
| Vals Legal Index | 69.2 | 75.9 | 68.4 | 75.3 | 1 | 0 of 4 | 7.5 |
| APEX-Agents 1.1 | 87.6 | 90.4 | 75.1 | 77.0 | 0 | 0 of 4 | 15.3 |
| BullshitBench V2 | 67.3 | 72.0 | 69.0 | 57.6 | 0 | 0 of 4 | 14.4 |
| CAIS Risk Indexscore, lower is better | 8.40 | 33.4 | 35.5 | 32.4 | 0 | 0 of 4 | 27.1 |
| CAIS Text Capabilities Indexscore | 66.3 | 59.0 | 63.9 | 47.6 | 0 | 0 of 4 | 18.7 |
| DesignArena Agentic OverallElo | 1,354 | 1,307 | 1,385 | 1,162 | 0 | 0 of 4 | 223 |
| DesignArena CodeElo | 1,374 | 1,331 | 1,396 | 1,394 | 0 | 0 of 4 | 65 |
| DesignArena Image to HTMLElo | 1,247 | 1,211 | 1,304 | 1,264 | 0 | 0 of 4 | 93 |
| DesignArena PDFElo | 1,296 | 1,335 | 1,194 | 1,200 | 0 | 0 of 4 | 141 |
| DesignArena UI ComponentElo | 1,357 | 1,349 | 1,401 | 1,368 | 0 | 0 of 4 | 52 |
| EBR-bench | 90.5 | 87.6 | 76.2 | 31.4 | 0 | 0 of 4 | 59.1 |
| Extended NYT Connections | 97.3 | 99.3 | 98.4 | 93.4 | 0 | 0 of 4 | 5.9 |
| Furniture Assembly Benchmark | 90.0 | 77.5 | 80.0 | 54.2 | 0 | 0 of 4 | 35.8 |
| GDPx | 16.4 | 10.9 | 9.60 | 6.82 | 0 | 0 of 4 | 9.6 |
| Mystery Game Puzzles | 90.0 | 73.0 | 84.0 | 51.0 | 0 | 0 of 4 | 39.0 |
| OTIS Mock AIME 2024-2025 | 99.7 | 100.0 | 100.0 | 99.4 | 0 | 0 of 4 | 0.6 |
| RWE-bench | 62.0 | 20.0 | 37.0 | 0.0 | 0 | 0 of 4 | 62.0 |
| SWE Atlas v1.1 | 80.5 | 71.4 | 69.4 | 64.8 | 0 | 0 of 4 | 15.7 |
| TaxCalcBench v2 | 82.0 | 71.4 | 70.0 | 83.7 | 0 | 0 of 4 | 13.7 |
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.