GPT-5.2 vs o3
GPT-5.2 Verdict
- GPT-5.2 and o3 are level on quality (61.6 vs 61.6).
- GPT-5.2 is stronger in composite, human preference, instruction following, knowledge, long context, multimodal.
- o3 is stronger in agents & tools, coding, reasoning, maths.
- o3 is 1.4× cheaper ($3.50 vs $4.81 per 1M blended).
- o3 streams 1.6× faster (107 vs 68 tokens per second).
| Metric | GPT-5.2 | o3 |
|---|---|---|
| BenchLeader Index | 61.6 | 61.6 |
| Agents & tools score | 61.2 | 70.3 |
| Coding score | 53.7 | 62.0 |
| Composite score | 67.5 | 54.5 |
| Human preference score | 67.8 | 62.3 |
| Instruction following score | 67.4 | 65.0 |
| Knowledge score | 61.1 | 58.0 |
| Long context score | 67.9 | 63.8 |
| Multimodal score | 59.9 | 57.9 |
| Reasoning score | 61.3 | 61.5 |
| Maths score | – | 78.6 |
| Blended price $/M | $4.81 | $3.50 |
| Output speed | 68 tok/s | 107 tok/s |
| Time to first answer | 129.4 s | 6.3 s |
| Context window | 400k | 200k |
| Humanity's Last Exam | 27.8% | – |
| Terminal-Bench | 64.9% | – |
| SimpleBench | 45.8% | – |
| Remote Labor Index | 2.1% | – |
| APEX-Agents | 23.0% | – |
| Epoch Capabilities Index | 153.5 | 146.9 |
| LMArena Text | 1476 | 1432 |
| LMArena Hard Prompts | 1497 | 1441 |
| LMArena Coding | 1515 | 1460 |
| LMArena WebDev | 1417 | – |
| LMArena Vision | 1268 | 1214 |
| AA Intelligence Index | 30.4 | 20.2 |
| IFBench | 75.4% | 71.4% |
| AA-LCR | 82.7% | 74.7% |
| MMMU-Pro | – | 70.1% |
| AA-Omniscience | -0.9 | -15.6 |
| Terminal-Bench Hard | 47.0% | 37.1% |
| GPQA Diamond (AA) | 90.3% | 82.7% |
| Humanity's Last Exam (AA) | 37.7% | 20.1% |
| τ²-Bench Telecom (AA) | 84.8% | 80.7% |
| SWE-Bench Pro | 29.9% | – |
| PRBench Finance | – | 47.7% |
| PRBench Legal | – | 48.6% |
| VISTA | 46.6% | – |
| MultiNRC | 42.2% | – |
| TutorBench | 53.5% | – |
| MMMU (validation) | – | 82.9% |
| MMMU-Pro (official) | – | 76.4% |
| Kagi LLM Benchmark | 73.3% | 67.6% |
| IFEval (HELM) | – | 86.9% |
| Omni-MATH (HELM) | – | 71.4% |
| WildBench (HELM) | – | 86.1% |
| MMLU-Pro (HELM) | – | 85.9% |
| GPQA Diamond (HELM) | – | 75.3% |
| HELM Capabilities mean | – | 81.1% |
| Aider Polyglot | – | 81.3% |
| SWE-bench Verified (bash only) | 69.0% | 58.4% |
| SWE-bench Verified (any scaffold) | – | 58.4% |
| ARC-AGI-1 | 94.5% | – |
| ARC-AGI-2 | 72.9% | – |
| BFCL Overall | 55.9% | 63.0% |
Data as of 2026-09-12. Best configuration of each model; every score links to its source on the model pages.
GPT-5.2 vs o3: questions
- Is GPT-5.2 better than o3?
- GPT-5.2 and o3 are level on quality (61.6 vs 61.6). The BenchLeader Index combines every independent quality benchmark; o3 is ahead overall as of 2026-09-12, but check the category scores for your use.
- Is GPT-5.2 better than o3 for coding?
- o3 scores higher in coding (62 vs 54 on the category index, where 50 is average).
- Is GPT-5.2 better than o3 for agentic tasks?
- o3 scores higher in agentic tasks (70 vs 61 on the category index, where 50 is average).
- Which is cheaper, GPT-5.2 or o3?
- o3 is cheaper: $3.50 against $4.81 per million tokens, blended at three input tokens per output token.
- Which is faster, GPT-5.2 or o3?
- o3 streams faster: 107 against 68 output tokens per second.
- Which has the larger context window?
- GPT-5.2 accepts more context: 400k against 200k tokens.