Anthropic·released 2024-06-201 source
Claude 3.5 Sonnet benchmark scores: 12 benchmarks tracked. 42% of its scores are independently reproduced — the rest are self-reported or unverified.
Leads on 0 of 12 benchmarks·best result 82 on MMLU (unverified)·5 of 12 independently reproduced
| Benchmark | Score | Evidence status | Measured |
|---|---|---|---|
| MMLU | 82accuracy (%) | unverified· optimizedT1 | 2024-06-20Stanford CRFM Leaderboard |
| MATH level 5 | 51.68accuracy (%) | reproduced· optimizedT1 | 2024-06-20Epoch AI |
| DTBench | 46.28accuracy (%) | unverifiedT1 | 2024-06-20https://conceptualreasoning.ai/dtbench |
| METR Time Horizons | 40.15accuracy (%) | self-reportedT1 | 2024-06-20https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/ |
| GPQA diamond | 38.72accuracy (%) | reproduced· optimizedT1 | 2024-06-20Epoch AI |
| WeirdML | 30.97accuracy (%) | unverifiedT1 | 2024-06-20https://htihle.github.io/weirdml.html |
| Cybench | 17.5accuracy (%) | unverified· optimizedT1 | 2024-06-20Cybench leaderboard |
| SimpleBench | 13accuracy (%) | unverifiedT1 | 2024-06-20SimpleBench Leaderboard |
| OTIS Mock AIME 2024-2025 | 6.43accuracy (%) | reproduced· optimizedT1 | 2024-06-20Epoch AI |
| FrontierMath-2025-02-28-Private | 1.81accuracy (%) | reproduced· optimizedT1 | 2024-06-20Epoch AI |
| FrontierMath-Tier-4-2025-07-01-Private | 0accuracy (%) | reproducedT1 | 2024-06-20Epoch AI |
| VPCT | 0accuracy (%) | unverifiedT1 | 2024-06-20VPCT leaderboard |
Claims drawn from cited facts, not live model generation.
This model was developed by Anthropic and released in 2024, placing it in a recent vintage of the developer's lineup. That origin and timing matter more than any single number: a 2024 release from Anthropic signals a model built on relatively current training and alignment practices, so its value lies in being contemporary rather than merely new.
4 cited facts
This model is scored on 12 tracked benchmarks. It holds no current top score on any of them. Its record can be trusted, since outside evaluation has independently reproduced its scores.
3 cited facts
On average across its benchmarks, this model trails the state-of-the-art leader by 56.22 points, a score gap measured under a disclosed harness rather than a statement about capability. A gap of that size reads as a wide one, placing the model well behind the front of the pack, consistent with its holding the top score on none of the benchmarks tracked (0 leader benchmarks).
3 cited facts
1 facts cross-checked across data sources: 1 single-source.
Source facts, citations, and refresh stamp for this record.
Sources: epoch_benchmarks
Claude 3.5 Sonnet is an AI model developed by Anthropic, released 2024-06-20. Its benchmark record below tags every score as independently reproduced, self-reported, or unverified.
Claude 3.5 Sonnet has recorded scores on 12 benchmarks, each shown with its evidence status.
Claude 3.5 Sonnet has recorded scores on 12 benchmarks — GPQA diamond, MATH level 5, OTIS Mock AIME 2024-2025, DTBench, WeirdML, MMLU, and 6 more. The full table above shows each score with its evidence status.
5 of 12 recorded scores (42%) are independently reproduced rather than self-reported by the lab.
Claude 3.5 Sonnet has 12 tracked claims: 5 independently reproduced, 1 self-reported, 6 unverified.