Gemini 3.5 Flash vs GPT-5.6 Terra
Gemini 3.5 Flash 
GPT-5.6 Terra Verdict
- Gemini 3.5 Flash (medium) and GPT-5.6 Terra (xhigh) are level on quality (63.9 vs 64.2).
- Gemini 3.5 Flash (medium) is stronger in human preference, instruction following, knowledge, multimodal, reasoning.
- GPT-5.6 Terra (xhigh) is stronger in agents & tools, coding, composite, long context, maths.
- Gemini 3.5 Flash (medium) is 1.3× cheaper ($3.38 vs $4.50 per 1M blended).
- Gemini 3.5 Flash (medium) streams 2.2× faster (209 vs 93 tokens per second).
| Metric | Gemini 3.5 Flash (medium) | GPT-5.6 Terra (xhigh) |
|---|---|---|
| BenchLeader Index | 63.9 | 64.2 |
| Agents & tools score | 67.8 | 69.6 |
| Coding score | 58.9 | 63.7 |
| Composite score | 71.4 | 77.1 |
| Human preference score | 67.4 | 66.3 |
| Instruction following score | 72.2 | 65.0 |
| Knowledge score | 73.9 | 60.9 |
| Long context score | 63.6 | 66.0 |
| Multimodal score | 67.6 | 63.0 |
| Reasoning score | 66.8 | 58.2 |
| Maths score | – | 71.0 |
| Blended price $/M | $3.38 | $4.50 |
| Output speed | 209 tok/s | 93 tok/s |
| Time to first answer | 17.0 s | 32.5 s |
| Context window | 1M | 1M |
| SimpleBench | – | 48.9% |
| SciCode | – | 51.6% |
| ProofBench | – | 74.0% |
| LMArena Text | 1476 | 1466 |
| LMArena Hard Prompts | 1493 | 1491 |
| LMArena Coding | 1503 | 1519 |
| LMArena WebDev | 1491 | 1521 |
| LMArena Vision | 1306 | 1269 |
| LMArena Agent | – | 1.5 |
| AA Intelligence Index | 33.6 | 38.2 |
| IFBench | 74.6% | 66.3% |
| AA-LCR | 74.3% | 79.0% |
| MMMU-Pro | 83.9% | 79.5% |
| AA-Omniscience | 20.8 | -3.0 |
| Terminal-Bench Hard | 39.4% | 62.9% |
| GPQA Diamond (AA) | 92.1% | 90.8% |
| Humanity's Last Exam (AA) | 41.3% | 41.9% |
| SciCode (AA) | – | 52.3% |
| τ²-Bench Telecom (AA) | 95.6% | 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.