Muse Spark vs Muse Wonder
As of 13 Oct 2026, Muse Wonder (max) is the better pick for most workloads: it leads on quality (66.3 vs 64.5 on the BenchLeader Index).
Muse Spark Key differences
- Muse Wonder (max) leads on quality: 66.3 vs 64.5.
- Muse Spark is stronger in agents & tools, coding, human preference, instruction following, maths, multimodal.
- Muse Wonder (max) is stronger in composite, knowledge, long context, reasoning.
Choose Muse Spark if
It does not lead on anything measured here; the other is the safer default.
Choose Muse Wonder (max) if
- reasoning is the priority (88 vs 66 on the category score)
- knowledge is the priority (73 vs 63 on the category score)
- long context is the priority (66 vs 64 on the category score)
- you need a longer context: 1.0M against 262k tokens
Category by category
Category scores on the BenchLeader scale: 50 is the average evaluated model, 15 points is one standard deviation. A gap under one point is called level.
| Category | Muse Spark | Muse Wonder (max) | Leader |
|---|---|---|---|
| Reasoning | 65.9 | 87.8 | Muse Wonder |
| Coding | 65.3 | – | not comparable |
| Agents & tools | 66.0 | – | not comparable |
| Maths | 59.3 | – | not comparable |
| Knowledge | 63.4 | 73.0 | Muse Wonder |
| Instruction following | 78.3 | – | not comparable |
| Human preference | 68.7 | – | not comparable |
| Multimodal | 64.6 | – | not comparable |
| Long context | 64.1 | 65.5 | Muse Wonder |
| Composite | 64.7 | 78.2 | Muse Wonder |
Price, speed and latency
List prices blended at three input tokens per output token. Per-task prices use the token counts on the methodology page; suite cost is what Artificial Analysis paid to run its whole Intelligence Index.
| Measure | Muse Spark | Muse Wonder (max) |
|---|---|---|
| Blended price per 1M tokens | – | $8.00 |
| Input / output per 1M | – / – | $4.00 / $20.00 |
| Price per task: chat reply | – | $0.0076 |
| Price per task: summarise a 30-page report | – | $0.060 |
| Price per task: code edit | – | $0.054 |
| Price per task: agentic coding session | – | $0.320 |
| Price per task: structured extraction | – | $0.012 |
| Cost to run a full benchmark suite | – | $4.52 |
| Output speed | – | 85 tok/s |
| Time to first answer | – | 60.8 s |
| Time to a complete answer | – | 66.7 s |
Every benchmark
| Metric | Muse Spark | Muse Wonder (max) |
|---|---|---|
| BenchLeader Index | 64.5 | 66.3 |
| Agents & tools score | 66.0 | – |
| Coding score | 65.3 | – |
| Composite score | 64.7 | 78.2 |
| Human preference score | 68.7 | – |
| Instruction following score | 78.3 | – |
| Knowledge score | 63.4 | 73.0 |
| Long context score | 64.1 | 65.5 |
| Maths score | 59.3 | – |
| Multimodal score | 64.6 | – |
| Reasoning score | 65.9 | 87.8 |
| Blended price $/M | – | $8.00 |
| Output speed | – | 85 tok/s |
| Time to first answer | – | 60.8 s |
| Context window | 262k | 1.0M |
| GPQA Diamond | 89.8% | – |
| OTIS Mock AIME | 88.9% | – |
| Humanity's Last Exam | 40.6% | – |
| SciCode | 51.5% | – |
| ProofBench | 17.0% | – |
| Epoch Capabilities Index | 152.0 | – |
| LMArena Text | 1489 | – |
| LMArena Hard Prompts | 1506 | – |
| LMArena Coding | 1529 | – |
| LMArena Vision | 1306 | – |
| LiveBench | – | 81.2% |
| LiveBench Reasoning | – | 90.6% |
| LiveBench Coding | – | 80.7% |
| LiveBench Agentic Coding | – | 62.7% |
| LiveBench Mathematics | – | 96.5% |
| LiveBench Data Analysis | – | 81.6% |
| LiveBench Language | – | 82.9% |
| LiveBench Instruction Following | – | 73.0% |
| AA Intelligence Index v4.3.2 | 31.3 | 52.4 |
| IFBench | 75.9% | – |
| AA-LCR | 78.0% | 80.7% |
| MMMU-Pro | 80.5% | – |
| AA-Omniscience | 7.2 | 25.5 |
| Terminal-Bench Hard | 45.5% | – |
| GPQA Diamond (AA) | 88.4% | – |
| Humanity's Last Exam (AA) | 40.7% | 50.1% |
| SciCode (AA) | – | 60.3% |
| τ²-Bench Telecom (AA) | 91.5% | – |
| AIME (Vals) | 96.9% | – |
| MMLU-Pro | 87.3% | – |
| LegalBench | 84.2% | – |
| CorpFin | 65.1% | – |
| TaxEval | 77.7% | – |
| SWE-bench (Vals) | 74.4% | – |
| GPQA Diamond (Vals) | 89.7% | – |
| SWE-Bench Pro | 55.0% | – |
| MCP Atlas | 82.2% | – |
| MultiChallenge | 75.5% | – |
| PRBench Finance | 52.4% | – |
| MultiNRC | 59.0% | – |
| TutorBench | 68.5% | – |
| CritPt | 11.3% | 28.3% |
| GDPval-AA v2.1 | 24.9% | 63.5% |
| CaseLaw v2 | 63.1% | – |
| MedCode | 51.3% | – |
| MedScribe | 85.9% | – |
| MMMU-Pro (Vals) | 87.4% | – |
| Terminal-Bench 2.0 (Vals) | 59.5% | – |
| Vibe Code Bench v1.1 | 19.7% | – |
| FORTRESS | 20.2% | – |
| SWE-Bench Pro (private) | 55.0% | – |
| SWE Atlas: Codebase QnA | 24.2% | – |
| SWE Atlas: Test Writing | 31.1% | – |
| LMArena Maths | 1466 | – |
| LMArena Creative Writing | 1465 | – |
| LMArena Instruction Following | 1464 | – |
| LMArena Multi-turn | 1493 | – |
| LMArena Longer Queries | 1476 | – |
| LMArena Document | 1467 | – |
| Terminal-Bench 4.0 (AA) | – | 52.0% |
| Terminal-Bench 2.1 (AA) | 62.2% | – |
| AutomationBench | – | 68.1% |
| GDP.pdf | – | 30.6% |
| AA-Omniscience: accuracy | 49.6% | 41.1% |
| AA-Omniscience: non-hallucination | 15.8% | 73.5% |
| AA-Briefcase v1.1 | – | 1659 |
Data as of 2026-10-13. Best configuration of each model; every score links to its source on the model pages.
Muse Spark vs Muse Wonder: questions
- Is Muse Spark better than Muse Wonder?
- Muse Wonder (max) leads on quality: 66.3 vs 64.5. The BenchLeader Index combines every independent quality benchmark; Muse Wonder (max) is ahead overall as of 2026-10-13, but check the category scores for your use.
- Which has the larger context window?
- Muse Wonder accepts more context: 1.0M against 262k tokens.