Up to four models
Gemini 3.8 Flash vs GPT-5.5
On the index (GPT-4 = 100), Gemini 3.8 Flash scores 173 and GPT-5.5 172. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
173157–189
WeightsClosed
ReleasedSep 2026
Price in / out$0.82 / $4.13
Context1.05M
Core tests taken7 of 9
GPT-5.5
172156–188
WeightsClosed
ReleasedApr 2026
Price in / out$5.0 / $30
Context1.05M
Core tests taken6 of 9
Gemini 3.8 Flash and GPT-5.5 are too close to call (1 point apart, within the uncertainty), ahead on 4 of 5 shared core tests, and costs 6.8× less.
Test by testthird-party results
Every result as a table
| Test | Gemini 3.8 Flash | GPT-5.5 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 95.4% | 94.0% |
| GPQA Diamond (Vals AI) | 94.4% | 93.2% |
| MMLU Pro | 90.2% | 88.1% |
| MMMU Pro | 89.1% | 88.3% |
| SimpleBench | 82.4% | 69.0% |
| SimpleQA Verified | 69.7% | 63.0% |
| Maths | ||
| FrontierMath (tiers 1-3) | 68.4% | 85.3% |
| AIME-style maths (OTIS mock) | 98.9% | 100.0% |
| Chess puzzles | 58.0% | 63.0% |
| FrontierMath (tier 4) | 21.9% | 72.5% |
| Surface Evolver Bench | 76.9% | 88.1% |
| Coding | ||
| Code Migration | 36.5% | 45.2% |
| LiveCodeBench | 89.5% | 85.3% |
| ProgramBench | 1.0% | 0.5% |
| SWE-bench | 80.0% | 82.6% |
| Vibe Code Bench v1.1 | 78.6% | 69.8% |
| DeepSWE | 73.8% | 67.0% |
| MirrorCode | 38.9% | 23.3% |
| WeirdML | 84.8% | 84.9% |
| Agents | ||
| APEX-Agents | 64.3% | 55.1% |
| SkillsBench | 58.0% | 62.2% |
| SRE Bench | 11.1% | 3.8% |
| DTBench | 95.7% | 96.0% |
| Furniture assembly | 31.7% | 44.2% |
| LMCA | 52.9% | 54.3% |
| Mystery game puzzles | 47.0% | 56.0% |
| Professional work | ||
| Finance Agent (v2) | 61.4% | 51.8% |
| EMB | 72.2% | 64.5% |
| LegalBench | 87.0% | 86.5% |
| Legal Research Bench | 38.9% | 40.4% |
| MedCode | 48.1% | 49.1% |
| MedScribe | 84.5% | 86.9% |
| MortgageTax | 65.3% | 68.8% |
| Public Benefits Bench v1.1 | 65.3% | 60.9% |
| Tax Agent Bench | 32.1% | 21.8% |
| TaxEval v2 | 74.5% | 75.0% |
| Novel problems | ||
| ARC-AGI-2 | 89.2% | 85.0% |
| ARC-AGI-1 | 98.5% | 95.0% |
| ARC-AGI-3 | 10.4% | 0.4% |
Tests at least two of these models have taken. Bold rows are the core tests.