Anthropic·released 2026-07-241 source
Claude Opus 5 benchmark scores: 23 benchmarks tracked, leading 1. 22% of its scores are independently reproduced — the rest are self-reported or unverified.
Leads on 1 of 23 benchmarks·best result 99 on ProofBench (unverified)·5 of 23 independently reproduced
Head-to-headClaude Opus 5 vs GPT-5.6 SolLeads on: OSWorld 2.0
| Benchmark | Score | Evidence status | Measured |
|---|---|---|---|
| ProofBench | 99accuracy (%) | unverifiedT1 | 2026-07-24https://www.vals.ai/benchmarks/proof_bench |
| OTIS Mock AIME 2024-2025 | 98.89accuracy (%) | reproduced· optimizedT1 | 2026-07-24Epoch AI |
| ARC-AGI | 97.5accuracy (%) | unverified· optimizedT1 | 2026-07-24https://arcprize.org/leaderboard |
| DTBench | 96.45accuracy (%) | unverifiedT1 | 2026-07-24https://conceptualreasoning.ai/dtbench |
| GPQA diamond | 91.84accuracy (%) | reproduced· optimizedT1 | 2026-07-24Epoch AI |
| WeirdML | 91.78accuracy (%) | unverifiedT1 | 2026-07-24https://htihle.github.io/weirdml.html |
| ARC-AGI-2 | 90.42accuracy (%) | unverifiedT2 | 2026-07-24 |
| FrontierMath-Tiers-1-3-v2-Private | 85.61accuracy (%) | reproducedT1 | 2026-07-24Epoch AI |
| SimpleBench | 76.72accuracy (%) | unverifiedT1 | 2026-07-24https://lmcouncil.ai/benchmarks |
| LMCA | 75.82accuracy (%) | unverifiedT1 | 2026-07-24https://conceptualreasoning.ai/lmca |
| DeepSWE | 73.65accuracy (%) | unverifiedT1 | 2026-07-24https://deepswe.datacurve.ai/ |
| FrontierMath-Tier-4-v2-Private | 73.17accuracy (%) | reproducedT1 | 2026-07-24Epoch AI |
| APEX-Agents | 65.8accuracy (%) | unverifiedT2 | 2026-07-24 |
| Balrog | 63.4accuracy (%) | unverifiedT1 | 2026-07-24https://balrogai.com/ |
| SimpleQA Verified | 59.9accuracy (%) | unverifiedT2 | 2026-07-24 |
| Mystery Game Puzzles | 54.84accuracy (%) | reproducedT1 | 2026-07-24Epoch AI |
| FrontierCode | 53.38accuracy (%) | unverifiedT1 | 2026-07-24https://cognition.com/frontiercode |
| FrontierSWE | 52.01accuracy (%) | unverifiedT1 | 2026-07-24https://www.frontierswe.com/ |
| EBR-bench | 45.71accuracy (%) | unverifiedT2 | 2026-07-24 |
| Furniture Assembly | 44.05accuracy (%) | unverifiedT2 | 2026-07-24 |
| Chess Puzzles | 40.03accuracy (%) | unverifiedT2 | 2026-07-24 |
| PostTrainBench | 35.04accuracy (%) | unverifiedT2 | 2026-07-24 |
| OSWorld 2.0 | 31.43accuracy (%) | unverifiedT1 | 2026-07-24https://osworld-v2.xlang.ai/ |
Claims drawn from cited facts, not live model generation.
This model was developed by Anthropic and released in 2026.
2 cited facts
This model has scores on 23 tracked benchmarks and currently leads 1 of them. The record’s evidence status is independently reproduced.
3 cited facts
Its mean gap to the best recorded score across its tracked benchmarks is 10.34 points.
1 cited fact
1 facts cross-checked across data sources: 1 single-source.
Source facts, citations, and refresh stamp for this record.
Sources: epoch_benchmarks
Claude Opus 5 is an AI model developed by Anthropic, released 2026-07-24. Its benchmark record below tags every score as independently reproduced, self-reported, or unverified.
Claude Opus 5 has recorded scores on 23 benchmarks, each shown with its evidence status.
Claude Opus 5 has recorded scores on 23 benchmarks — GPQA diamond, OTIS Mock AIME 2024-2025, DTBench, WeirdML, LMCA, SimpleQA Verified, and 17 more. The full table above shows each score with its evidence status.
5 of 23 recorded scores (22%) are independently reproduced rather than self-reported by the lab.
Claude Opus 5 has 23 tracked claims: 5 independently reproduced, 18 unverified.