Up to four models
Gemini 3.8 Flash vs Gemini 3.7 Flash
On the index (GPT-4 = 100), Gemini 3.8 Flash scores 175 and Gemini 3.7 Flash 174. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
Gemini 3.8 Flash175159–191
WeightsClosed
ReleasedSep 2026
Price in / out$0.82 / $4.13
Context1.05M
Core tests taken7 of 9
Gemini 3.7 Flash174158–191
WeightsClosed
ReleasedAug 2026
Price in / out$0.75 / $3.75
Context1.05M
Core tests taken6 of 9
Gemini 3.8 Flash and Gemini 3.7 Flash are too close to call (1 point apart, within the uncertainty), ahead on 4 of 6 shared core tests, at a similar price.
Test by testthird-party results
Gemini 3.8 FlashGemini 3.7 FlashEvery result as a table
| Test | Gemini 3.8 Flash | Gemini 3.7 Flash |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 95.4% | 94.8% |
| GPQA Diamond (Vals AI) | 94.4% | 93.9% |
| MMLU Pro | 90.2% | 90.1% |
| MMMU Pro | 89.1% | 89.0% |
| SimpleQA Verified | 69.7% | 69.2% |
| Maths | ||
| FrontierMath (tiers 1-3) | 68.4% | 71.6% |
| AIME-style maths (OTIS mock) | 98.9% | 97.2% |
| ProofBench v1.1 | 48.0% | 58.0% |
| Chess puzzles | 58.0% | 66.0% |
| FrontierMath (tier 4) | 21.9% | 36.6% |
| Coding | ||
| Terminal-Bench 4.0 | 19.7% | 12.1% |
| Code Migration | 36.5% | 34.8% |
| CyberBench v1.1 | 43.8% | 43.8% |
| IOI | 56.9% | 67.8% |
| LiveCodeBench | 89.5% | 88.6% |
| ProgramBench | 1.0% | 0.0% |
| SWE-bench | 80.0% | 80.8% |
| Vibe Code Bench v1.1 | 78.6% | 70.4% |
| DeepSWE | 73.8% | 65.5% |
| FrontierSWE | 19.6% | 20.3% |
| Agents | ||
| APEX-Agents | 64.3% | 67.8% |
| SkillsBench | 58.0% | 65.9% |
| SRE Bench | 11.1% | 4.6% |
| Terminal-Bench-Science | 8.6% | 7.1% |
| DTBench | 95.7% | 96.8% |
| Furniture assembly | 31.7% | 26.7% |
| LMCA | 52.9% | 50.4% |
| Mystery game puzzles | 47.0% | 37.0% |
| Professional work | ||
| Finance Agent (v2) | 61.4% | 59.0% |
| EMB | 72.2% | 71.3% |
| LegalBench | 87.0% | 87.3% |
| Legal Research Bench | 38.9% | 34.6% |
| MedCode | 48.1% | 53.4% |
| MedScribe | 84.5% | 83.9% |
| MortgageTax | 65.3% | 66.6% |
| SAGE | 35.1% | 49.2% |
| Tax Agent Bench | 32.1% | 22.8% |
| TaxEval v2 | 74.5% | 74.7% |
| Novel problems | ||
| ARC-AGI-2 | 89.2% | 84.6% |
| ARC-AGI-1 | 98.5% | 95.5% |
Tests at least two of these models have taken. Bold rows are the core tests.
