- Humanity's Last Exam (no tools), Preview Thinking (09-2025) 7.3
- GPQA diamond, Thinking 71.7
- AIME 2025, Thinking (non-thinking 50.1) 48.2
- LiveCodeBench v5 (UI: 1/1/2025-5/1/2025), Thinking 58.4
- SWE-Bench Verified single attempt, Thinking (non-thinking 41.3) 38.9
- SimpleQA Verified, Thinking 9.6
- FACTS Grounding, Thinking 87.5
- MMMU single attempt, Thinking 72
- Vibe-Eval (Reka), Thinking 59.8
- MRCR v2 (8-needle) 128k (average), Thinking (1M pointwise 7.7) 25.6
- Global MMLU (Lite), Thinking 84.9
Gemini 2.5 Flash-Lite Sep 2025
Gemini 2.5 Flash-Lite places in the top three on 3 benchmarks.
47
Benchmarks with a score
0
Ranked first, out of five or more models
3
In the top three
Google DeepMind Sep 2025 Gemini 2.5 Flash-Lite
Sep 2025 snapshot Closed weights
128.2
#287 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
- LiveCodeBench v5 58.4% #31 /47
- AV-SpeakerBench 47.2% #3 /5
- CorpFin v2 57.6% #82 /119
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Vectara Hallucination Leaderboard (2025 dataset) | 3.30% | #2 of 104 | Official leaderboard | |
| FACTS Grounding | 87.5% | #4 of 15 | Lab model card | |
| HELM Arabic | 79.5 | #10 of 28 | Official leaderboard | |
| Global MMLU Lite | 84.9% | #7 of 16 | Lab model card | |
| LMArena (Elo) | 1,379 | #154 of 335 | Official leaderboard | |
| HELM Arabic Enterprise | 62.6 | #16 of 23 | Official leaderboard | |
| IFEval | 81% | #42 of 60 | Official leaderboard | |
| Phare | 58.9% | #52 of 62 | Official leaderboard | |
| Judgemark v2.1 | 56.0 | #36 of 42 | Official leaderboard | |
| MRCR v2 | 25.6% | #28 of 30 | Lab model card | |
| Humanity's Last Exam | 7.30% | #148 of 154 | Lab model card | |
| SimpleQA Verified | 9.60% | #22 of 22 | Lab model card |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GPQA Diamond | 71.7% | #168 of 302 | Lab model card | |
| GPQA Main | 30.9% | #55 of 60 | Official leaderboard | |
| AIME 2025 | 48.2% | #133 of 143 | Lab model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveCodeBench v5 | 58.4% | #31 of 47 | Lab model card | |
| DesignArena Website | 1,092 | #118 of 151 | Official leaderboard | |
| DesignArena Code | 1,071 | #117 of 143 | Official leaderboard | |
| DesignArena Game Dev | 1,037 | #119 of 143 | Official leaderboard | |
| DesignArena UI Component | 1,016 | #119 of 139 | Official leaderboard | |
| DesignArena 3D Design | 948 | #126 of 136 | Official leaderboard | |
| SWE-bench Verified | 38.9% | #146 of 155 | Lab model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BFCL V4 | 36.9% | #33 of 65 | Official leaderboard | |
| SnakeBench | 23 | #103 of 188 | Official leaderboard | |
| PinchBench | 11.8% | #48 of 48 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LegalBench | 82% | #75 of 138 | Official leaderboard | |
| CorpFin v2 | 57.6% | #82 of 119 | Official leaderboard | |
| MedScribe | 75.8% | #72 of 97 | Official leaderboard | |
| MedCode | 34.2% | #75 of 96 | Official leaderboard | |
| MortgageTax | 57.5% | #69 of 87 | Official leaderboard | |
| TaxEval v2 | 66.2% | #104 of 129 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| DECIDE | 52.3 | #2 of 8 | Official leaderboard | |
| MSEB | 17.1 | #4 of 9 | Official leaderboard | |
| Daily-Omni | 61.9% | #7 of 14 | Official leaderboard | |
| MMMU | 72% | #56 of 103 | Lab model card | |
| AV-SpeakerBench | 47.2% | #3 of 5 | Official leaderboard | |
| RoboRewardBench | 0.99 | #14 of 22 | Official leaderboard | |
| Vision Arena | 1,174 | #91 of 136 | Official leaderboard | |
| Vibe-Eval | 59.8% | #7 of 9 | Lab model card | |
| VPCT | 30% | #27 of 27 | VPCT leaderboard, via Epoch AI |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AIR-Bench 2024 | 65.8% | #42 of 82 | Official leaderboard | |
| HELM Safety | 91.0 | #40 of 75 | Official leaderboard | |
| Anthropic Red Team | 98.7% | #45 of 75 | Official leaderboard | |
| SimpleSafetyTests | 96.5% | #61 of 75 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| HELM Capabilities | 59.1 | #44 of 60 | Official leaderboard | |
| LMCA | 18.1% | #117 of 147 | Official leaderboard | |
| SAGE (Vals) | 30.8% | #74 of 81 | Official leaderboard |
Its lab's numbers put it at 128.2, number 287 of 404
The Self-Reported Index fits one score to the 11 capability results Google DeepMind reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 125.3 to 131.1. How we compute it.
- FACTS Grounding 87.5
- Global MMLU Lite 84.9
- MMMU 72
- GPQA Diamond 71.7
- Vibe-Eval 59.8
- LiveCodeBench v5 58.4
- AIME 2025 48.2
- SWE-bench Verified 38.9
- MRCR v2 25.6
- SimpleQA Verified 9.6
- Humanity's Last Exam 7.3
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.