Meta · Benchmarks

Muse Spark Open benchmarks

Muse Spark Open has 3 published scores in this dataset, from a Meta release dated 2026-10-15. Its composite index is 88.0, which places it 9th of 22 ranked models and 5th among 15 open-weight models. Its strongest area is reasoning (index 92.7) and its weakest is coding (index 84.3). Its best placing is on Humanity's Last Exam (with tools), where its 61.0 ranks 7th of 17 models with a published score.

Open weightsOfficial page

Category profile

Index 0–100: each score relative to the best published score on that test.

ReasoningCodingAgenticKnowledge
Reasoning
92.71 test · not ranked
Coding
84.31 test · not ranked
Agentic
87.11 test · not ranked
Knowledge
—no data

Model facts

Lab
Meta
Type
Open weights
Released
2026-10-15
Parameters
816B total, 42B active (MoE)
Context
1M
License
Muse Model License (manual-approval gate)

Muse Spark Open hardware requirements and cost

Even at 4-bit it needs about 563 GB (4× H200 (564 GB)), well beyond a home workstation, so most people run it on rented multi-GPU servers. Full BF16 precision needs about 1.9 TB (Multiple nodes (over 1.1 TB)).

Estimated GPU memory for Muse Spark Open
PrecisionWeightsWith headroomSmallest setup
BF16 (full precision)1.6 TB1.9 TBMultiple nodes (over 1.1 TB)
FP8 / 8-bit816 GB938 GB8× H200 (1,128 GB)
4-bit (GGUF Q4_K_M)490 GB563 GB4× H200 (564 GB)

Estimate: parameters × bytes per parameter, plus 15% for the KV cache and runtime at modest context lengths. Long contexts and large batches need more.

Lighter or stronger open models

Open models ranked just above and below on the composite index, with their 4-bit memory needs.

Muse Spark Open speed and pricing

Measured figures link to their source; estimates are labelled.

API output speed
196 tok/sArtificial Analysis median · Artificial Analysis
Time to first token
1.18 sOpenRouter model page provider stats
API price per 1M tokens
$1.25 in · $4.25 outcheapest provider on OpenRouter: Fireworks · OpenRouter endpoints API
Compare speed and price across all models

Score by test

Rank is among every model with a published score on that test.

GPQA Diamond89.0%xhigh effort#14 of 19 · 7.1 behind GPT-6 AstraHumanity's Last Exam (with tools)61.0%with tools, xhigh effort#7 of 17 · 9.0 behind Kimi K3.1 ProTerminal-Bench76.4%Terminal-Bench 2.1, Terminus 2 agent, xhigh effort#10 of 13 · 14.2 behind DeepSeek V4.1 Flash

Nearby in the composite ranking

Models with the closest composite index.

  1. 1DeepSeek V4 Pro91.0
  2. 2Kimi K390.6
  3. 3Tencent Hy4 Preview89.9
  4. 4Muse Spark Open88.0
  5. 5Reflection Beam 1.087.0
  6. 6GPT-6.1 Sol86.5
  7. 7Qwen3.6 27B85.8

Head-to-head comparisons

Side-by-side pages built from the tests both models report.

No head-to-head page for Muse Spark Open yet. Compare it on the rankings dashboard.

Sources

ModelTestScoreSettingSource
Muse Spark OpenGPQA Diamond89.0xhigh effortHugging Face model card (Muse-Spark-Open-816B-A42B)
Muse Spark OpenHumanity's Last Exam (with tools)61.0with tools, xhigh effortHugging Face model card (Muse-Spark-Open-816B-A42B)
Muse Spark OpenTerminal-Bench76.4Terminal-Bench 2.1, Terminus 2 agent, xhigh effortHugging Face model card (Muse-Spark-Open-816B-A42B)

Scores are published results from model cards, lab reports, and public leaderboards, each linked to its source. Settings such as thinking mode or test version differ between sources and are shown next to each score. OpenLLM Buddy catalog models are uncensored fine-tunes; their scores are those of the official base model. Data collected 2026-10-30. How the index works