GPT-5.4 vs GPT-5.5
GPT-5.4 Verdict
- GPT-5.5 leads on quality: 66.6 vs 63.4.
- GPT-5.4 is stronger in composite.
- GPT-5.5 is stronger in agents & tools, coding, human preference, instruction following, knowledge, long context, multimodal, reasoning.
- GPT-5.4 is 2.0× cheaper ($5.63 vs $11.25 per 1M blended).
- GPT-5.4 streams 1.5× faster (130 vs 88 tokens per second).
| Metric | GPT-5.4 | GPT-5.5 |
|---|---|---|
| BenchLeader Index | 63.4 | 66.6 |
| Agents & tools score | 66.1 | 68.4 |
| Coding score | 51.6 | 57.3 |
| Composite score | 78.1 | 77.7 |
| Human preference score | 64.5 | 67.6 |
| Instruction following score | 71.7 | 73.3 |
| Knowledge score | 66.8 | 73.7 |
| Long context score | 67.6 | 68.8 |
| Multimodal score | 64.0 | 64.9 |
| Reasoning score | 62.3 | 69.7 |
| Blended price $/M | $5.63 | $11.25 |
| Output speed | 130 tok/s | 88 tok/s |
| Time to first answer | 126.4 s | 62.3 s |
| Context window | 1.1M | 1.1M |
| Terminal-Bench | 81.8% | 84.7% |
| SimpleBench | – | 69.0% |
| Remote Labor Index | – | 6.3% |
| APEX-Agents | 34.9% | 38.5% |
| FrontierCode | – | 43.0% |
| Epoch Capabilities Index | 156.9 | 159.1 |
| LMArena Text | 1466 | 1477 |
| LMArena Hard Prompts | 1488 | 1498 |
| LMArena Coding | 1514 | 1509 |
| LMArena WebDev | 1386 | 1458 |
| LMArena Vision | 1293 | 1296 |
| LMArena Agent | – | 3 |
| AA Intelligence Index | 39.0 | 38.6 |
| IFBench | 74.0% | 75.8% |
| AA-LCR | 82.0% | 84.3% |
| MMMU-Pro | 78.4% | 79.9% |
| AA-Omniscience | 5.8 | 20.5 |
| Terminal-Bench Hard | 57.6% | 60.6% |
| GPQA Diamond (AA) | 92.0% | 93.5% |
| Humanity's Last Exam (AA) | 43.7% | 45.8% |
| SciCode (AA) | – | 55.8% |
| τ²-Bench Telecom (AA) | 87.1% | 93.9% |
| HiL-Bench | 9.7% | 39.7% |
| EQ-Bench 4 | 1272 | 1315 |
| Kagi LLM Benchmark | 63.8% | 88.8% |
Data as of 2026-09-09. Best configuration of each model; every score links to its source on the model pages.