- Humanity's Last Exam (no tools), Thinking (06-17) 6.9
- GPQA diamond, Thinking 66.7
- AIME 2025, Thinking 63.1
- LiveCodeBench v5 (UI: 1/1/2025-5/1/2025), Thinking 34.3
- Aider Polyglot (whole), Thinking 27.1
- SWE-Bench Verified, multiple attempts (single attempt 27.6), Thinking 44.9
- SimpleQA, Thinking 13
- FACTS Grounding, Thinking 86.8
- MMMU single attempt, Thinking 72.9
- Vibe-Eval (Reka), Thinking 57.5
- MRCR v2 (8-needle) 128k (average), Thinking (1M pointwise 5.40) 30.6
- Global MMLU (Lite), Thinking 84.5
Gemini 2.5 Flash-Lite Jun 2025
Gemini 2.5 Flash-Lite.
Google DeepMind Released 17 Jun 2025 30
Benchmarks with a score
0
Ranked first, out of five or more models
0
In the top three
Google DeepMind Jun 2025 Gemini 2.5 Flash-Lite
Jun 2025 snapshot Closed weights
133.8
#151 of 271
Epoch Capabilities Index
- LiveCodeBench v5 34.3% #38 /44
- ProfBench 44.3% #49 /72
- StructEval 72% #18 /28
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 | |
|---|---|---|---|---|
| FACTS Grounding | 86.8% | #5 of 15 | Model card | |
| LMArena (Elo) | 1,375 | #155 of 333 | Official leaderboard | |
| Global MMLU Lite | 84.5% | #8 of 16 | Model card | |
| Wer wird Millionär? LLM Benchmark | 700.0 | #50 of 76 | Official leaderboard | |
| Vectara Hallucination Leaderboard (HHEM) | 2.90% | #110 of 148 | Official leaderboard | |
| SimpleQA | 13% | #44 of 59 | Model card | |
| Fiction.LiveBench | 47.2% | #44 of 57 | Epoch AI | |
| MRCR v2 | 30.6% | #26 of 27 | Model card | |
| Humanity's Last Exam | 6.90% | #143 of 147 | Model card |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GPQA Diamond | 66.7% | #185 of 284 | Model card | |
| DTBench | 38% | #129 of 178 | Official leaderboard | |
| AIME 2025 | 63.1% | #123 of 143 | Model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| StructEval | 72% | #18 of 28 | Official leaderboard | |
| WeirdML | 35.2% | #93 of 112 | Official leaderboard | |
| LiveCodeBench v5 | 34.3% | #38 of 44 | Model card | |
| SWE-bench Verified | 44.9% | #133 of 150 | Model card | |
| ALE-Bench | 326 | #111 of 121 | Epoch AI | |
| Aider Polyglot | 27.1% | #27 of 29 | Model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Galileo Agent Leaderboard v2 | 47% | #8 of 21 | Official leaderboard | |
| LisanBench | 122.3 | #111 of 133 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ProfBench | 44.3% | #49 of 72 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MMMU | 72.9% | #48 of 97 | Model card | |
| Vision Arena | 1,189 | #86 of 136 | Official leaderboard | |
| Vibe-Eval | 57.5% | #8 of 9 | Model card |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LMCA | 18.1% | #109 of 139 | Epoch AI | |
| Conceptual Reasoning Index | 25.1 | #125 of 149 | Official leaderboard | |
| CAIS Vision Capabilities Index | 38.1 | #46 of 53 | Official leaderboard | |
| ACCoRD | 33.4 | #138 of 150 | Official leaderboard | |
| CAIS Text Capabilities Index | 9.20 | #62 of 66 | Official leaderboard | |
| CAIS Risk Index | 77.1 | #65 of 65 | Official leaderboard |
Google DeepMind's models over time
Google DeepMind's models in color, against every other model on the Epoch Capabilities Index, by release date.
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.