Muse Spark Open (xhigh): benchmark scores, price and ranking
Muse Spark Open (xhigh) from Meta AI ranks #20 of 747 LLMs on the LLMs Tiger aggregate benchmark score (54.7), based on 8 benchmark results captured up to 2026-10-21.
- Organisation
- Meta AI
- Type
- Open-weight / local model
- Open weights
- Yes
- Released
- 2026-10-15 (source: epoch)
- Cheapest API price
- $1.25 in / $4.25 out per million tokens (via baseten, observed 2026-10-21T00:47Z)
- Aggregate rank
- #20 of 747 (score 54.7)
Relative to other models, Muse Spark Open (xhigh) is strongest on SciCode (#12 of 109) and weakest on FrontierMath T4 (#27 of 54).
Across the 8 benchmarks where at least ten models are scored, Muse Spark Open (xhigh) is in the top 10% on 0 and in the top half on 8.
Its cheapest listed input price, $1.25 per million tokens, is lower than that of 27% of the 429 models with a listed price.
Muse Spark Open (xhigh) benchmark results
| Benchmark | Score | Rank | Source | Captured |
|---|---|---|---|---|
| SciCode | 58.4% | #12 of 109 | Epoch AI Benchmarking Hub | 2026-10-21 |
| OTIS Mock AIME | 98.1% | #23 of 193 | Epoch AI Benchmarking Hub | 2026-10-21 |
| ALE-Bench | 1,590 | #10 of 83 | Epoch AI Benchmarking Hub | 2026-10-21 |
| GPQA Diamond | 88.6% | #46 of 281 | Epoch AI Benchmarking Hub | 2026-10-21 |
| SimpleQA Verified | 56.6% | #13 of 70 | Epoch AI Benchmarking Hub | 2026-10-21 |
| CritPt | 14.9% | #32 of 118 | Epoch AI Benchmarking Hub | 2026-10-21 |
| FrontierMath T1-3 | 69.8% | #23 of 72 | Epoch AI Benchmarking Hub | 2026-10-21 |
| FrontierMath T4 | 31.7% | #27 of 54 | Epoch AI Benchmarking Hub | 2026-10-21 |
About Muse Spark Open (xhigh)
Muse Spark Open (xhigh) is an open-weight model that can run on your own hardware from Meta AI. It was released on 2026-10-15, according to Epoch AI. The cheapest API listing we track charges $1.25 per million input tokens and $4.25 per million output tokens, through baseten. We hold 8 benchmark results for it across 3 areas, and none is in the top tenth of its board.
Where Muse Spark Open (xhigh) stands
On our aggregate score, Muse Spark Open (xhigh) is #20 of 747 ranked models, in the top fifty. Its score is 54.7; the scale is centred on 50, so it sits 4.7 points above the typical model. That beats 97% of the ranked field. Among the 29 ranked Meta AI models it is #2. The next one up is Muse Wonder (max) at 56.9. The next one down is Muse Spark 1.3 (max) at 53.9.
Muse Spark Open (xhigh), benchmark by benchmark
SciCode puts Muse Spark Open (xhigh) at 58.4%, #12 of 109. That beats 89% of the models tested, but 8.4 points short of Claude Opus 5.5 (max) (66.9%). The results either side are Muse Spark 1.3 (max) (58.8%) and GPT-6 Sol (max) (57.6%). It covers scientific research code. The board median is 45.1%, so Muse Spark Open (xhigh) is 13.3 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-21.
On OTIS Mock AIME, Muse Spark Open (xhigh) reaches 98.1%. It ranks #23 of 193, ahead of 88% of the field, and 1.9 points behind Claude Fable 5.1 (max) at 100.0%. The results either side are Grok 4.7 (xhigh) (98.1%) and Grok 4.5 (high) (97.8%). It covers competition maths (AIME 2024-25 style). The board median is 59.2%, so Muse Spark Open (xhigh) is 38.9 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-21.
Muse Spark Open (xhigh) manages 1,590 on ALE-Bench, good for #10 of 83. GPT-6 Astra (max) leads at 2,951, 1,361 points further up, and 88% of models sit below. The results either side are GPT-5.3 Codex (xhigh) (1,655) and Kimi K3 (max) (1,524). It covers AtCoder heuristic-optimisation contests (performance rating). The board median is 785, so Muse Spark Open (xhigh) is 805 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-21.
GPQA Diamond puts Muse Spark Open (xhigh) at 88.6%, #46 of 281. That beats 84% of the models tested, but 7.8 points short of gemini-4-argon-high (96.4%). The results either side are grok-4.3 (high) (88.8%) and Inkling Small (xhigh) (88.5%). It covers expert-level science questions. The board median is 72.6%, so Muse Spark Open (xhigh) is 16.0 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-21.
On SimpleQA Verified, Muse Spark Open (xhigh) reaches 56.6%. It ranks #13 of 70, ahead of 81% of the field, and 19.4 points behind gemini-4-argon-high at 76.0%. The results either side are Claude Opus 5 (max) (59.9%) and Grok 4.7 (xhigh) (56.0%). It covers short factual questions. The board median is 41.0%, so Muse Spark Open (xhigh) is 15.6 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-21.
Muse Spark Open (xhigh) manages 14.9% on CritPt, good for #32 of 118. GPT-5.6 Sol (max) leads at 32.3%, 17.4 points further up, and 73% of models sit below. The results either side are Muse Spark 1.1 (xhigh) (15.1%) and deepseek-v4.1-flash-max (14.3%). It covers unpublished research-level physics problems. The board median is 3.1%, so Muse Spark Open (xhigh) is 11.7 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-21.
FrontierMath T1-3 puts Muse Spark Open (xhigh) at 69.8%, #23 of 72. That beats 68% of the models tested, but 23.9 points short of GPT-6 Astra (max) (93.7%). The results either side are Claude Opus 4.7 (max) (70.2%) and GLM-5.3 (max) (68.8%). It covers research-level maths, tiers 1-3. The board median is 57.2%, so Muse Spark Open (xhigh) is 12.6 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-21.
On FrontierMath T4, Muse Spark Open (xhigh) reaches 31.7%. It ranks #27 of 54, ahead of 50% of the field, and 68.3 points behind GPT-6.1 Sol (max) at 100.0%. The results either side are Grok 4.6 (xhigh) (31.7%) and GPT-5.2 (xhigh) (31.7%). It covers research-level maths, tier 4. The board median is 31.7%, so Muse Spark Open (xhigh) is 0.0 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-21.
Strengths and weaknesses
Software engineering: strong, ahead of 88% of the field on average across 2 benchmarks. Its best board here is SciCode (#12 of 109) and its weakest is ALE-Bench (#10 of 83). Reasoning, maths and science: solid, ahead of 73% of the field on average across 5 benchmarks. Its best board here is OTIS Mock AIME (#23 of 193) and its weakest is FrontierMath T4 (#27 of 54).
Price and value
The median input price across the 429 priced models is $0.500 per million tokens, so Muse Spark Open (xhigh) costs more than the typical listing. Models that score within 1.5 points or better and cost less: GLM-5.3 (max) ($0.079, 53.6), Claude Haiku 5.5 (max) ($0.100, 54.8), GPT-6 Luna (max) ($0.100, 53.4). Similar scores cost more on GPT-5.3 Codex (xhigh) ($1.75) and gemini-3.1-pro-preview-high ($2.00).
Release history
The Meta AI model released before it was Muse Wonder (max) on 2026-10-12, which scores 56.9.
Reading these numbers
The 8 scores on this page come from 1 source: Epoch AI Benchmarking Hub (8). No official leaderboard supplied any of them, so treat them as third-party measurements. The leaderboard also lists 2 other reasoning-effort settings of this model, which redirect here. This page shows the highest.
Muse Spark Open (xhigh) is also listed at 2 other reasoning-effort settings; this page shows the highest-effort row, which is the one the leaderboard keeps by default.
Models ranked nearby
- #17 gemini-3.1-pro-preview-high (55.3)
- #18 GPT-5.6 Luna (max) (54.8)
- #19 Claude Haiku 5.5 (max) (54.8)
- #21 Claude 4 Sonnet (54.5)
- #22 Claude 4.5 Opus (54.5)
- #23 Claude 3.5-Sonnet-20241022 (54.4)
More from Meta AI
- Muse Wonder (max) (#13, 56.9)
- Muse Spark 1.3 (max) (#29, 53.9)
- Muse Spark (#34, 53.6)
- Muse Spark 1.1 (xhigh) (#68, 52.4)
- Llama 3.1-405B (#130, 51.4)
- Muse Spark 1.2 (xhigh) (#133, 51.3)
Compare it on the live leaderboard → · How the aggregate score works