Gemini 3.5 Flash-Lite.
Google DeepMind Jul 2026 Closed weights Small
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.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LMArena (Elo) | 1,455 | #58 of 335 | Official leaderboard | |
| BullshitBench V2 | 66% | #26 of 129 | Official leaderboard | |
| BullshitBench | 52.7% | #27 of 107 | Official leaderboard | |
| Creative Writing v3 | 72.3 | #42 of 120 | Official leaderboard | |
| Last Translation Benchmark | 11.9 | #14 of 36 | Official leaderboard | |
| StudyArena Text | 1,236 | #39 of 85 | Official leaderboard | |
| SimpleQA Verified | 33.1% | #27 of 56 | Official leaderboard | |
| MRCR v2 | 72.2% | #15 of 30 | Lab model card | |
| FACTS Parametric | 43.4% | #23 of 45 | Official leaderboard | |
| Global MMLU Lite | 89.5% | #22 of 43 | Official leaderboard | |
| OpenAI MRCR v2 | 40.3% | #27 of 47 | Official leaderboard | |
| FACTS Grounding v2 | 66% | #15 of 25 | Official leaderboard | |
| Extended NYT Connections | 60.4% | #65 of 103 | Official leaderboard | |
| Hemingway-bench | 990 | #29 of 42 | Official leaderboard | |
| Antidote | 991 | #25 of 36 | Official leaderboard | |
| FACTS Parametric V2 | 43% | #19 of 27 | Official leaderboard | |
| FACTS Benchmark Suite | 53.5% | #20 of 23 | Official leaderboard | |
| FACTS Search V2 | 63.4% | #20 of 21 | Official leaderboard | |
| FACTS Multimodal | 36.1% | #23 of 23 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GPQA Diamond | 83.3% | #89 of 229 | Epoch AI run | |
| DTBench | 83.5% | #75 of 180 | Official leaderboard | |
| OTIS Mock AIME 2024-2025 | 71.1% | #108 of 212 | Epoch AI run | |
| ARC-AGI-2 | 10.3% | #59 of 98 | Official leaderboard | |
| ProofBench | 13% | #40 of 59 | Official leaderboard | |
| ARC-AGI | 53.5% | #66 of 95 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 26% | #70 of 86 | Official leaderboard | |
| Human Behavior Bench | 75.9 | #9 of 11 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 0% | #68 of 68 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| IOI (Vals) | 26.2% | #18 of 55 | Official leaderboard | |
| Chart Arena | 1,547 | #5 of 15 | Official leaderboard | |
| WebDev Arena | 1,447 | #55 of 105 | arena.ai, via Epoch AI | |
| Code Arena | 1,435 | #75 of 128 | Official leaderboard | |
| MLE-bench | 39.2% | #6 of 10 | Lab model card | |
| Vibe Code Bench | 37.2% | #75 of 120 | Official leaderboard | |
| SWE-Bench Pro | 54.2% | #56 of 81 | Lab model card | |
| ALE-Bench | 1,137 | #89 of 126 | Official leaderboard | |
| WeirdML | 39% | #83 of 112 | Official leaderboard | |
| Image-to-WebDev Arena | 1,449 | #43 of 55 | Official leaderboard | |
| APEX-SWE | 31.2% | #51 of 65 | Official leaderboard | |
| AndroidBench | 50.6% | #28 of 32 | Official leaderboard | |
| Code Migration | 6.15% | #70 of 78 | Official leaderboard | |
| SWE-sweep | 0.10% | #10 of 10 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| OSWorld-Verified | 74% | #17 of 41 | Lab model card | |
| Toolathlon-Verified | 57.1% | #15 of 25 | Official leaderboard | |
| Mystery Game Puzzles | 19% | #54 of 84 | Epoch AI run | |
| RuneBench | 4.18 | #53 of 76 | Official leaderboard | |
| Chess Puzzles | 21% | #68 of 93 | Epoch AI run | |
| AutomationBench | 3.96% | #35 of 42 | Official leaderboard | |
| Terminal-Bench 2.1 | 54% | #49 of 56 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MortgageTax | 68.7% | #13 of 87 | Official leaderboard | |
| CorpBench Work | 87.3 | #5 of 17 | Official leaderboard | |
| LegalBench | 84.1% | #41 of 138 | Official leaderboard | |
| ProfBench | 51.6% | #26 of 72 | Official leaderboard | |
| MedCode | 43.5% | #42 of 96 | Official leaderboard | |
| TaxEval v2 | 72.6% | #59 of 129 | Official leaderboard | |
| CorpFin v2 | 60.7% | #65 of 119 | Official leaderboard | |
| PolicyBench | 81.5% | #29 of 50 | Official leaderboard | |
| Vals Finance Agent v2 | 47.4% | #53 of 80 | Official leaderboard | |
| Vals Index v2 | 36.7% | #48 of 65 | Official leaderboard | |
| Excel Modeling Benchmark | 43.2% | #60 of 76 | Official leaderboard | |
| MedScribe | 70.9% | #83 of 97 | Official leaderboard | |
| Legal Research Bench | 13.9% | #80 of 93 | Official leaderboard | |
| MedXpertQA MM Extended | 47.7% | #36 of 41 | Official leaderboard | |
| Public Benefits Bench | 51.8% | #44 of 50 | Official leaderboard | |
| APEX-1 | 54.7% | #63 of 71 | Official leaderboard | |
| GDPval-AA v2 | 1,140 | #24 of 27 | Lab model card | |
| Tax Agent Bench | 7.25% | #75 of 84 | Official leaderboard | |
| APEX-Agents | 29.3% | #56 of 61 | Collected by Epoch AI, source not named | |
| APEX-Agents 1.1 | 46.1% | #60 of 65 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 0% | #77 of 80 | Official leaderboard | |
| APEX-Accounting | 5.20% | #49 of 49 | Official leaderboard | |
| Immigration Bench | 55.7% | #25 of 25 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Vision Arena | 1,266 | #31 of 136 | Official leaderboard | |
| ZeroBench | 14% | #30 of 76 | Official leaderboard | |
| 1H-VideoQA | 75.3% | #10 of 20 | Official leaderboard | |
| CharXiv | 76.5% | #47 of 75 | Lab model card | |
| ParseBench | 59.5% | #28 of 44 | Official leaderboard | |
| MineBench | 1,339 | #48 of 75 | Official leaderboard | |
| vCX Benchmark | 82.7% | #31 of 42 | Official leaderboard | |
| Chartography | 10.2% | #40 of 43 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| nocot-bench (NCRI) | 80.6 | #58 of 185 | Official leaderboard | |
| CyberBench (Vals) | 72.3% | #11 of 26 | Official leaderboard | |
| CyberBench v1.1 (Vals) | 59.1% | #34 of 44 | Official leaderboard | |
| BrowseSafe-Adapt | 93.5% | #5 of 6 | Official leaderboard | |
| DeepsecBench | 2.13 | #29 of 29 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ACCoRD | 68.0 | #60 of 152 | Official leaderboard | |
| SAGE (Vals) | 47.3% | #32 of 81 | Official leaderboard | |
| LMCA | 37.6% | #66 of 147 | Official leaderboard | |
| Conceptual Reasoning Index | 50.7 | #68 of 151 | Official leaderboard | |
| SAGE v1.1 (Vals) | 79.5% | #26 of 29 | Official leaderboard |
The Self-Reported Index fits one score to the 5 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 144.2 to 151.3. How we compute it.
Google DeepMind's models in color, against every other model on the Epoch Capabilities Index, by release date.
Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)
The benchmarks the lab chose to headline, with the scores it gave.
Every result in the longer documents the lab published with it. 5 of them aren't in the launch post.