Gemini 3.8 Flash vs GPT-5.6 Terra
Gemini 3.8 Flash 
GPT-5.6 Terra Verdict
- Gemini 3.8 Flash (medium) and GPT-5.6 Terra (xhigh) are level on quality (63.5 vs 64.2).
- Gemini 3.8 Flash (medium) is stronger in composite, knowledge, long context, multimodal.
- GPT-5.6 Terra (xhigh) is stronger in coding, agents & tools, human preference, instruction following, maths, reasoning.
- Gemini 3.8 Flash (medium) is 3.0× cheaper ($1.50 vs $4.50 per 1M blended).
- Gemini 3.8 Flash (medium) streams 2.9× faster (273 vs 93 tokens per second).
| Metric | Gemini 3.8 Flash (medium) | GPT-5.6 Terra (xhigh) |
|---|---|---|
| BenchLeader Index | 63.5 | 64.2 |
| Coding score | 63.6 | 63.7 |
| Composite score | 79.4 | 77.1 |
| Knowledge score | 77.6 | 60.9 |
| Long context score | 68.7 | 66.0 |
| Multimodal score | 68.9 | 63.0 |
| Agents & tools score | – | 69.6 |
| Human preference score | – | 66.3 |
| Instruction following score | – | 65.0 |
| Maths score | – | 71.0 |
| Reasoning score | – | 58.2 |
| Blended price $/M | $1.50 | $4.50 |
| Output speed | 273 tok/s | 93 tok/s |
| Time to first answer | 13.8 s | 32.5 s |
| Context window | 1M | 1M |
| SimpleBench | – | 48.9% |
| SciCode | 54.4% | 51.6% |
| ProofBench | – | 74.0% |
| LMArena Text | – | 1466 |
| LMArena Hard Prompts | – | 1491 |
| LMArena Coding | – | 1519 |
| LMArena WebDev | – | 1521 |
| LMArena Vision | – | 1269 |
| LMArena Agent | – | 1.5 |
| AA Intelligence Index | 40 | 38.2 |
| IFBench | – | 66.3% |
| AA-LCR | 84.0% | 79.0% |
| MMMU-Pro | 84.2% | 79.5% |
| AA-Omniscience | 28.6 | -3.0 |
| Terminal-Bench Hard | – | 62.9% |
| GPQA Diamond (AA) | 93.5% | 90.8% |
| Humanity's Last Exam (AA) | 42.1% | 41.9% |
| SciCode (AA) | 55.1% | 52.3% |
| τ²-Bench Telecom (AA) | – | 80.4% |
| LiveCodeBench | – | 85.9% |
| MMLU-Pro | – | 86.7% |
| IOI | – | 65.3% |
| CorpFin | – | 65.3% |
| TaxEval | – | 76.2% |
| GPQA Diamond (Vals) | – | 90.9% |
| ARC-AGI-1 | – | 94.0% |
| ARC-AGI-2 | – | 74.2% |
| ARC-AGI-3 | – | 0.7% |
Data as of 2026-09-09. Best configuration of each model; every score links to its source on the model pages.