GPT-5.4 vs GPT-5.6 Sol
GPT-5.4 Verdict
- GPT-5.6 Sol (high) leads on quality: 68.0 vs 63.4.
- GPT-5.4 is stronger in human preference, instruction following, long context.
- GPT-5.6 Sol (high) is stronger in agents & tools, coding, composite, knowledge, multimodal, reasoning.
- GPT-5.4 is 1.4× cheaper ($5.63 vs $8.00 per 1M blended).
- GPT-5.4 streams 2.2× faster (130 vs 59 tokens per second).
| Metric | GPT-5.4 | GPT-5.6 Sol (high) |
|---|---|---|
| BenchLeader Index | 63.4 | 68.0 |
| Agents & tools score | 66.1 | 87.6 |
| Coding score | 51.6 | 72.0 |
| Composite score | 78.1 | 82.5 |
| Human preference score | 64.5 | – |
| Instruction following score | 71.7 | 67.6 |
| Knowledge score | 66.8 | 73.7 |
| Long context score | 67.6 | 67.4 |
| Multimodal score | 64.0 | 66.5 |
| Reasoning score | 62.3 | 62.9 |
| Blended price $/M | $5.63 | $8.00 |
| Output speed | 130 tok/s | 59 tok/s |
| Time to first answer | 126.4 s | 31.4 s |
| Context window | 1.1M | 1M |
| Terminal-Bench | 81.8% | – |
| SciCode | – | 56.9% |
| WeirdML | – | 88.8% |
| APEX-Agents | 34.9% | – |
| Epoch Capabilities Index | 156.9 | – |
| LMArena Text | 1466 | – |
| LMArena Hard Prompts | 1488 | – |
| LMArena Coding | 1514 | – |
| LMArena WebDev | 1386 | – |
| LMArena Vision | 1293 | – |
| AA Intelligence Index | 39.0 | 42.5 |
| IFBench | 74.0% | 69.2% |
| AA-LCR | 82.0% | 81.7% |
| MMMU-Pro | 78.4% | 81.8% |
| AA-Omniscience | 5.8 | 20.4 |
| Terminal-Bench Hard | 57.6% | 62.1% |
| GPQA Diamond (AA) | 92.0% | 92.8% |
| Humanity's Last Exam (AA) | 43.7% | 46.0% |
| SciCode (AA) | – | 57.8% |
| τ²-Bench Telecom (AA) | 87.1% | 83.3% |
| EnigmaEval | – | 37.1% |
| HiL-Bench | 9.7% | – |
| EQ-Bench 4 | 1272 | – |
| Kagi LLM Benchmark | 63.8% | – |
| ARC-AGI-1 | – | 97.0% |
| ARC-AGI-2 | – | 85.4% |
| ARC-AGI-3 | – | 2.2% |
Data as of 2026-09-09. Best configuration of each model; every score links to its source on the model pages.