Google DeepMind·released 2025-03-121 source
Gemma 3 27B benchmark scores: 10 benchmarks tracked. 20% of its scores are independently reproduced — the rest are self-reported or unverified.
Leads on 0 of 10 benchmarks·best result 79.9 on Lech Mazur Writing (unverified)·2 of 10 independently reproduced
| Benchmark | Score | Evidence status | Measured |
|---|---|---|---|
| Lech Mazur Writing | 79.9accuracy (%) | unverifiedT1 | 2025-03-12lechmazur/writing Github repository |
| MATH level 5 | 74.04accuracy (%) | reproduced· optimizedT1 | 2025-03-12Epoch AI |
| GeoBench | 52accuracy (%) | unverifiedT1 | 2025-03-12GeoBench leaderboard |
| Fiction.LiveBench | 33.3accuracy (%) | unverifiedT1 | 2025-03-11Fiction.live leaderboard |
| GPQA diamond | 30.3accuracy (%) | unverifiedT2 | 2025-03-12 |
| OTIS Mock AIME 2024-2025 | 22.42accuracy (%) | unverifiedT2 | 2025-03-12 |
| DTBench | 20.88accuracy (%) | unverifiedT1 | 2025-03-12https://conceptualreasoning.ai/dtbench |
| LMCA | 14.45accuracy (%) | unverifiedT1 | 2025-03-12https://conceptualreasoning.ai/lmca |
| Aider polyglot | 4.9accuracy (%) | unverified· optimizedT1 | 2025-03-12https://aider.chat/docs/leaderboards/ |
| Chess Puzzles | 0accuracy (%) | reproducedT1 | 2025-03-12Epoch AI |
Claims drawn from cited facts, not live model generation.
This model, developed by Google DeepMind, is a mid-size 27 billion parameter model released in 2025. This mid-size scale implies more capability headroom than compact models but higher serving cost than smaller ones.
4 cited facts
This model is scored on 10 tracked benchmarks, and it currently holds no top score on any of them. Its record has been independently reproduced, meaning outside evaluation backs the scores up.
3 cited facts
Across its benchmarks, this model trails the leading score by an average of 56.18 points, a wide gap that places it well behind the front of the pack rather than merely a couple of points off the frontier. It currently holds the top score on none of its benchmarks, so there is no category in which it can claim genuine leadership.
2 cited facts
2 facts cross-checked across data sources: 2 single-source.
Source facts, citations, and refresh stamp for this record.
Sources: curated_param_counts, epoch_benchmarks
Gemma 3 27B is an AI model developed by Google DeepMind, released 2025-03-12. Its benchmark record below tags every score as independently reproduced, self-reported, or unverified.
Gemma 3 27B has recorded scores on 10 benchmarks, each shown with its evidence status.
Gemma 3 27B has recorded scores on 10 benchmarks — Lech Mazur Writing, GPQA diamond, MATH level 5, OTIS Mock AIME 2024-2025, DTBench, LMCA, and 4 more. The full table above shows each score with its evidence status.
2 of 10 recorded scores (20%) are independently reproduced rather than self-reported by the lab.
Gemma 3 27B has 10 tracked claims: 2 independently reproduced, 8 unverified.