Google DeepMind·released 2026-08-131 source
Gemini 3.7 Flash benchmark scores: 18 benchmarks tracked. 33% of its scores are independently reproduced — the rest are self-reported or unverified.
Leads on 0 of 18 benchmarks·best result 97.22 on OTIS Mock AIME 2024-2025 (reproduced)·6 of 18 independently reproduced
| Benchmark | Score | Evidence status | Measured |
|---|---|---|---|
| OTIS Mock AIME 2024-2025 | 97.22accuracy (%) | reproduced· optimizedT1 | 2026-08-13Epoch AI |
| ARC-AGI | 95.5accuracy (%) | unverified· optimizedT1 | 2026-08-13https://arcprize.org/leaderboard |
| DTBench | 94.67accuracy (%) | unverifiedT1 | 2026-08-13https://conceptualreasoning.ai/dtbench |
| GPQA diamond | 93.1accuracy (%) | reproduced· optimizedT1 | 2026-08-13Epoch AI |
| ARC-AGI-2 | 84.58accuracy (%) | unverifiedT2 | 2026-08-13 |
| FrontierMath-Tiers-1-3-v2-Private | 71.58accuracy (%) | reproducedT1 | 2026-08-13Epoch AI |
| SimpleQA Verified | 69.2accuracy (%) | unverifiedT2 | 2026-08-13 |
| APEX-Agents | 67.8accuracy (%) | unverifiedT2 | 2026-08-13 |
| DeepSWE | 65.49accuracy (%) | unverifiedT1 | 2026-08-13https://deepswe.datacurve.ai/ |
| LMCA | 59.27accuracy (%) | unverifiedT1 | 2026-08-13https://conceptualreasoning.ai/lmca |
| ProofBench | 58accuracy (%) | unverifiedT1 | 2026-08-13https://www.vals.ai/benchmarks/proof_bench |
| Chess Puzzles | 44.23accuracy (%) | reproducedT1 | 2026-08-13Epoch AI |
| FrontierCode | 43.59accuracy (%) | unverifiedT1 | 2026-08-13https://cognition.com/frontiercode |
| FrontierMath-Tier-4-v2-Private | 36.59accuracy (%) | reproducedT1 | 2026-08-13Epoch AI |
| Mystery Game Puzzles | 30.6accuracy (%) | reproducedT1 | 2026-08-13Epoch AI |
| FrontierSWE | 20.26accuracy (%) | unverifiedT1 | 2026-08-13https://www.frontierswe.com/ |
| Remote Labor Index | 5accuracy (%) | unverifiedT2 | 2026-08-13 |
| Furniture Assembly | 0accuracy (%) | unverifiedT2 | 2026-08-13 |
Claims drawn from cited facts, not live model generation.
This model originates from Google DeepMind, placing it within the lineage of a major research organization rather than an independent or academic effort, so its design choices reflect the priorities and resources of a large industrial lab. Its vintage is 2026, which makes it a recent arrival whose capabilities and training assumptions sit close to the current state of the field rather than reflecting an earlier generation of techniques.
2 cited facts
This model is evaluated across 18 tracked benchmarks, yet it holds no current top score on any of them. Its record can be trusted as independently reproduced, meaning outside evaluation backs the reported scores up.
3 cited facts
Its mean gap to the best recorded score across its tracked benchmarks is 23.85 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
Gemini 3.7 Flash is an AI model developed by Google DeepMind, released 2026-08-13. Its benchmark record below tags every score as independently reproduced, self-reported, or unverified.
Gemini 3.7 Flash has recorded scores on 18 benchmarks, each shown with its evidence status.
Gemini 3.7 Flash has recorded scores on 18 benchmarks — GPQA diamond, OTIS Mock AIME 2024-2025, DTBench, LMCA, SimpleQA Verified, Chess Puzzles, and 12 more. The full table above shows each score with its evidence status.
6 of 18 recorded scores (33%) are independently reproduced rather than self-reported by the lab.
Gemini 3.7 Flash has 18 tracked claims: 6 independently reproduced, 12 unverified.