Google DeepMind·released 2025-05-061 source
Gemini 2.5 Pro (May 2025) benchmark scores: 9 benchmarks tracked. 22% of its scores are independently reproduced — the rest are self-reported or unverified.
Leads on 0 of 9 benchmarks·best result 95.9 on MATH level 5 (reproduced)·2 of 9 independently reproduced
| Benchmark | Score | Evidence status | Measured |
|---|---|---|---|
| MATH level 5 | 95.9accuracy (%) | reproduced· optimizedT1 | 2025-05-06Epoch AI |
| GeoBench | 86accuracy (%) | unverifiedT1 | 2025-05-06GeoBench leaderboard |
| Lech Mazur Writing | 80.9accuracy (%) | unverifiedT1 | 2025-05-06lechmazur/writing Github repository |
| Aider polyglot | 76.9accuracy (%) | unverified· optimizedT1 | 2025-05-06https://aider.chat/docs/leaderboards/ |
| Fiction.LiveBench | 66.7accuracy (%) | unverifiedT1 | 2025-05-06Fiction.live leaderboard |
| GPQA diamond | 55.56accuracy (%) | reproduced· optimizedT1 | 2025-05-06Epoch AI |
| The Agent Company | 30.3accuracy (%) | unverifiedT1 | 2025-05-06TheAgentCompany leaderboard |
| HLE | 13.66accuracy (%) | unverifiedT2 | 2025-05-06 |
| VPCT | 10.75accuracy (%) | unverifiedT1 | 2025-05-06VPCT leaderboard |
Claims drawn from cited facts, not live model generation.
This model originates from Google DeepMind and was released in 2025, placing it in the most recent wave of model development.
2 cited facts
This model is evaluated on 9 tracked benchmarks, but it does not currently hold the top score on any of them. Its benchmark results have been independently reproduced by outside evaluation, so the reported scores can be trusted as verified.
3 cited facts
Across its benchmarks the model trails the leading score by an average of 23.14 points under the disclosed harness — a score gap, not a capability verdict — and a margin that wide reads as well behind the front of the pack. It holds the top score on none of those benchmarks (sota leader count of 0), so there is no category leadership to point to yet.
2 cited facts
1 facts cross-checked across data sources: 1 single-source.
Source facts, citations, and refresh stamp for this record.
Sources: epoch_benchmarks
Gemini 2.5 Pro (May 2025) is an AI model developed by Google DeepMind, released 2025-05-06. Its benchmark record below tags every score as independently reproduced, self-reported, or unverified.
Gemini 2.5 Pro (May 2025) has recorded scores on 9 benchmarks, each shown with its evidence status.
Gemini 2.5 Pro (May 2025) has recorded scores on 9 benchmarks — Lech Mazur Writing, GPQA diamond, MATH level 5, Aider polyglot, The Agent Company, GeoBench, and 3 more. The full table above shows each score with its evidence status.
2 of 9 recorded scores (22%) are independently reproduced rather than self-reported by the lab.
Gemini 2.5 Pro (May 2025) has 9 tracked claims: 2 independently reproduced, 7 unverified.