GPT-5.4 vs Muse Spark 1.3
GPT-5.4 
Muse Spark 1.3 Verdict
- GPT-5.4 and Muse Spark 1.3 (xhigh) are level on quality (63.4 vs 63.9).
- GPT-5.4 is stronger in agents & tools, human preference, instruction following, reasoning.
- Muse Spark 1.3 (xhigh) is stronger in coding, composite, knowledge, long context, multimodal.
- Muse Spark 1.3 (xhigh) is 2.8× cheaper ($2.00 vs $5.63 per 1M blended).
- Muse Spark 1.3 (xhigh) streams 1.5× faster (193 vs 130 tokens per second).
| Metric | GPT-5.4 | Muse Spark 1.3 (xhigh) |
|---|---|---|
| BenchLeader Index | 63.4 | 63.9 |
| Agents & tools score | 66.1 | 60.4 |
| Coding score | 51.6 | 70.8 |
| Composite score | 78.1 | 78.6 |
| Human preference score | 64.5 | – |
| Instruction following score | 71.7 | – |
| Knowledge score | 66.8 | 75.0 |
| Long context score | 67.6 | 68.1 |
| Multimodal score | 64.0 | 66.6 |
| Reasoning score | 62.3 | – |
| Blended price $/M | $5.63 | $2.00 |
| Output speed | 130 tok/s | 193 tok/s |
| Time to first answer | 126.4 s | 40.0 s |
| Context window | 1.1M | 1M |
| Terminal-Bench | 81.8% | – |
| APEX-Agents | 34.9% | – |
| Epoch Capabilities Index | 156.9 | – |
| LMArena Text | 1466 | – |
| LMArena Hard Prompts | 1488 | – |
| LMArena Coding | 1514 | – |
| LMArena WebDev | 1386 | 1625 |
| LMArena Vision | 1293 | – |
| LiveBench | – | 81.6% |
| LiveBench Reasoning | – | 89.7% |
| LiveBench Coding | – | 81.1% |
| LiveBench Agentic Coding | – | 64.1% |
| LiveBench Mathematics | – | 96.0% |
| LiveBench Data Analysis | – | 79.6% |
| LiveBench Language | – | 82.8% |
| AA Intelligence Index | 39.0 | 45.2 |
| IFBench | 74.0% | – |
| AA-LCR | 82.0% | 83.0% |
| MMMU-Pro | 78.4% | 82.0% |
| AA-Omniscience | 5.8 | 23.1 |
| Terminal-Bench Hard | 57.6% | – |
| GPQA Diamond (AA) | 92.0% | 94.1% |
| Humanity's Last Exam (AA) | 43.7% | 47.5% |
| SciCode (AA) | – | 59.7% |
| τ²-Bench Telecom (AA) | 87.1% | – |
| Terminal-Bench 2.1 (Vals) | – | 72.3% |
| Vals Index | – | 60.3 |
| HiL-Bench | 9.7% | – |
| EQ-Bench 4 | 1272 | – |
| Kagi LLM Benchmark | 63.8% | – |
Data as of 2026-09-09. Best configuration of each model; every score links to its source on the model pages.