GPT-5 vs GPT-5.2-Codex
GPT-5 Verdict
- GPT-5 and GPT-5.2-Codex are level on quality (60.8 vs 60.8).
- GPT-5 is stronger in coding, composite, human preference, maths, reasoning.
- GPT-5.2-Codex is stronger in agents & tools, instruction following, knowledge, long context, multimodal.
- GPT-5 is 1.4× cheaper ($3.44 vs $4.81 per 1M blended).
- GPT-5 streams 3.8× faster (85 vs 22 tokens per second).
| Metric | GPT-5 | GPT-5.2-Codex |
|---|---|---|
| BenchLeader Index | 60.8 | 60.8 |
| Agents & tools score | 51.0 | 63.5 |
| Coding score | 60.9 | 55.7 |
| Composite score | 58.1 | 56.8 |
| Human preference score | 61.7 | – |
| Instruction following score | 65.0 | 75.1 |
| Knowledge score | 62.2 | 63.2 |
| Long context score | 65.6 | 67.7 |
| Maths score | 72.8 | – |
| Multimodal score | 59.6 | 60.8 |
| Reasoning score | 64.2 | – |
| Blended price $/M | $3.44 | $4.81 |
| Output speed | 85 tok/s | 22 tok/s |
| Time to first answer | 60.7 s | 5.9 s |
| Context window | 400k | 400k |
| Terminal-Bench | 49.6% | 66.5% |
| SciCode | 42.9% | – |
| Remote Labor Index | 1.7% | – |
| WeirdML | 39.8% | – |
| APEX-Agents | 18.3% | 27.6% |
| Epoch Capabilities Index | 150 | – |
| LMArena Text | 1427 | – |
| LMArena Hard Prompts | 1449 | – |
| LMArena Coding | 1462 | – |
| LMArena WebDev | – | 1339 |
| LMArena Vision | 1232 | – |
| 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 | 23.0 | 28.5 |
| IFBench | 73.1% | 77.6% |
| AA-LCR | 78.2% | 82.3% |
| MMMU-Pro | 74.2% | 76.3% |
| AA-Omniscience | -8.7 | -2.2 |
| Terminal-Bench Hard | 32.6% | 37.1% |
| GPQA Diamond (AA) | 85.3% | 89.9% |
| Humanity's Last Exam (AA) | 28.5% | 35.7% |
| τ²-Bench Telecom (AA) | 84.8% | 92.1% |
| LiveCodeBench | – | 88.0% |
| SWE-Bench Pro | – | 41.0% |
| PRBench Finance | 51.3% | – |
| PRBench Legal | 49.0% | – |
| VISTA | 49.7% | – |
| MultiNRC | 52.1% | – |
| TutorBench | 55.3% | – |
| Kagi LLM Benchmark | 72.7% | – |
| IFEval (HELM) | 87.5% | – |
| Omni-MATH (HELM) | 64.7% | – |
| WildBench (HELM) | 85.7% | – |
| MMLU-Pro (HELM) | 86.3% | – |
| GPQA Diamond (HELM) | 79.1% | – |
| HELM Capabilities mean | 80.7% | – |
| Aider Polyglot | 88.0% | – |
| SWE-bench Verified (bash only) | – | 72.8% |
| SWE-bench Verified (any scaffold) | 75.6% | 72.8% |
Data as of 2026-09-12. Best configuration of each model; every score links to its source on the model pages.
GPT-5 vs GPT-5.2-Codex: questions
- Is GPT-5 better than GPT-5.2-Codex?
- GPT-5 and GPT-5.2-Codex are level on quality (60.8 vs 60.8). The BenchLeader Index combines every independent quality benchmark; GPT-5.2-Codex is ahead overall as of 2026-09-12, but check the category scores for your use.
- Is GPT-5 better than GPT-5.2-Codex for coding?
- GPT-5 scores higher in coding (61 vs 56 on the category index, where 50 is average).
- Is GPT-5 better than GPT-5.2-Codex for agentic tasks?
- GPT-5.2-Codex scores higher in agentic tasks (64 vs 51 on the category index, where 50 is average).
- Which is cheaper, GPT-5 or GPT-5.2-Codex?
- GPT-5 is cheaper: $3.44 against $4.81 per million tokens, blended at three input tokens per output token.
- Which is faster, GPT-5 or GPT-5.2-Codex?
- GPT-5 streams faster: 85 against 22 output tokens per second.
- Which has the larger context window?
- Both accept 400k tokens of context.