April 2026
Muse Spark Benchmark Results
Full evaluation results across reasoning, science, medical, coding, and retrieval benchmarks
Last updated
May 27, 2026
Primary source
Official Meta AI announcementInterpretation
Strong on health, charts, and science; weaker on coding and abstract reasoning.
| Category | Benchmark | Muse Spark | GPT-5.4 | Gemini 3.1 Pro | Notes | Source |
|---|---|---|---|---|---|---|
| Overall | AA v4.0 | 52 | 57 | 57 | Comprehensive index | Artificial Analysis |
| Chart Understanding | CharXiv | 86.4 | 82.8 | 80.2 | Chart comprehension | Meta / AA |
| Medical | HealthBench Hard | 42.8 | 40.1 | 20.6 | Medical QA | Meta / HealthBench |
| Deep Search | DeepSearchQA | 74.8 | — | 69.7 | Research retrieval | Meta |
| Reasoning | HLE (Fast) | 36.5 | 43.9 | 48.4 | — | Meta |
| Reasoning | HLE (Contemplating) | 50.2 | 43.9 | 48.4 | Extended reasoning | Meta |
| Science | FrontierScience | 38.3 | 36.7 | 23.3 | Research frontier | Meta / FrontierScience |
| Abstract | ARC AGI 2 | 42.5 | 76.1 | 76.5 | Pattern reasoning | Meta / AA |
| Coding | Terminal-Bench 2.0 | 59.0 | 75.1 | 68.5 | Agentic coding | Meta / Terminal-Bench |
Sources combine Meta's launch materials and public benchmark summaries. Treat all launch-week benchmark data as subject to revision as third-party evaluators publish updated results.
86.4
CharXiv
#1 Chart Understanding
42.8
HealthBench Hard
#1 Medical QA
50.2
HLE (Contemplating)
#1 Extended Reasoning
Muse Spark leads on chart understanding, medical reasoning, deep search, science, and extended reasoning.
Try Muse Spark at meta.ai →