Gemini 3.1 Pro vs GPT-5.2-Codex
Gemini 3.1 Pro 
GPT-5.2-Codex Verdict
- Gemini 3.1 Pro leads on quality: 64.0 vs 60.8.
- Gemini 3.1 Pro is stronger in agents & tools, coding, composite, human preference, knowledge, maths, multimodal, reasoning.
- GPT-5.2-Codex is stronger in instruction following, long context.
- They cost about the same ($4.50 per 1M blended).
- Gemini 3.1 Pro streams 4.3× faster (108 vs 25 tokens per second).
| Metric | Gemini 3.1 Pro | GPT-5.2-Codex |
|---|---|---|
| BenchLeader Index | 64.0 | 60.8 |
| Agents & tools score | 63.7 | 63.5 |
| Coding score | 59.9 | 55.7 |
| Composite score | 67.4 | 56.8 |
| Human preference score | 59.9 | – |
| Instruction following score | 69.6 | 75.1 |
| Knowledge score | 66.2 | 63.1 |
| Long context score | 67.6 | 67.8 |
| Maths score | 60.7 | – |
| Multimodal score | 66.2 | 60.7 |
| Reasoning score | 69.8 | – |
| Blended price $/M | $4.50 | $4.81 |
| Output speed | 108 tok/s | 25 tok/s |
| Time to first answer | 30.5 s | 6.7 s |
| Context window | 1.0M | 400k |
| GPQA Diamond | 94.1% | – |
| FrontierMath Tiers 1–3 | 59.6% | – |
| FrontierMath Tier 4 | 26.8% | – |
| OTIS Mock AIME | 95.6% | – |
| Humanity's Last Exam | 46.4% | – |
| Terminal-Bench | 80.2% | 66.5% |
| SimpleBench | 79.6% | – |
| SciCode | 58.9% | – |
| WeirdML | 72.1% | – |
| APEX-Agents | 33.5% | 27.6% |
| ProofBench | 26.0% | – |
| GSO-Bench | 22.6% | – |
| Epoch Capabilities Index | 155 | – |
| LMArena Text | 1487 | – |
| LMArena Hard Prompts | 1507 | – |
| LMArena Coding | 1521 | – |
| LMArena WebDev | 1447 | 1339 |
| LMArena Vision | 1295 | – |
| LMArena Agent | -5.3 | – |
| LiveBench | – | 74.0% |
| LiveBench Reasoning | – | 77.7% |
| LiveBench Coding | – | 83.6% |
| LiveBench Agentic Coding | – | 49.4% |
| LiveBench Mathematics | – | 88.8% |
| LiveBench Data Analysis | – | 78.2% |
| LiveBench Language | – | 73.7% |
| AA Intelligence Index | 30.4 | 28.5 |
| IFBench | 77.1% | 77.6% |
| AA-LCR | 82.0% | 82.3% |
| MMMU-Pro | 82.4% | 76.3% |
| AA-Omniscience | 31.9 | -2.2 |
| Terminal-Bench Hard | 53.8% | 37.1% |
| GPQA Diamond (AA) | 94.1% | 89.9% |
| Humanity's Last Exam (AA) | 47.0% | 35.7% |
| SciCode (AA) | 58.7% | – |
| τ²-Bench Telecom (AA) | 95.6% | 92.1% |
| LiveCodeBench | – | 88.0% |
| AIME 2026 | 98.3% | – |
| HMMT February 2026 | 94.7% | – |
| MathArena Apex | 60.9% | – |
| SWE-Bench Pro | – | 41.0% |
| MultiChallenge | 71.4% | – |
| PRBench Finance | 41.9% | – |
| PRBench Legal | 44.0% | – |
| MultiNRC | 64.7% | – |
| HiL-Bench | 35.3% | – |
| TutorBench | 53.0% | – |
| EQ-Bench 4 | 1142 | – |
| SWE-bench Verified (bash only) | – | 72.8% |
| SWE-bench Verified (any scaffold) | – | 72.8% |
| ARC-AGI-1 | 98.0% | – |
| ARC-AGI-2 | 77.1% | – |
| ARC-AGI-3 | 0.4% | – |
Data as of 2026-09-10. Best configuration of each model; every score links to its source on the model pages.
Gemini 3.1 Pro vs GPT-5.2-Codex: questions
- Is Gemini 3.1 Pro better than GPT-5.2-Codex?
- Gemini 3.1 Pro leads on quality: 64.0 vs 60.8. The BenchLeader Index combines every independent quality benchmark; Gemini 3.1 Pro is ahead overall as of 2026-09-10, but check the category scores for your use.
- Is Gemini 3.1 Pro better than GPT-5.2-Codex for coding?
- Gemini 3.1 Pro scores higher in coding (60 vs 56 on the category index, where 50 is average).
- Is Gemini 3.1 Pro better than GPT-5.2-Codex for agentic tasks?
- Gemini 3.1 Pro scores higher in agentic tasks (64 vs 64 on the category index, where 50 is average).
- Which is cheaper, Gemini 3.1 Pro or GPT-5.2-Codex?
- Gemini 3.1 Pro is cheaper: $4.50 against $4.81 per million tokens, blended at three input tokens per output token.
- Which is faster, Gemini 3.1 Pro or GPT-5.2-Codex?
- Gemini 3.1 Pro streams faster: 108 against 25 output tokens per second.
- Which has the larger context window?
- Gemini 3.1 Pro accepts more context: 1.0M against 400k tokens.