Muse Glimmer-30BvsMuse Spark 1.1
Across 5 shared benchmarks, Muse Spark 1.1 leads overall: Muse Glimmer-30B wins 0, Muse Spark 1.1 wins 5, with 0 ties and an average score difference of -21.24.
Muse Glimmer-30B
Facebook AI研究实验室 · 2026-08-10 · Reasoning model
Muse Spark 1.1
Facebook AI研究实验室 · 2026-07-09 · Reasoning model
Muse Glimmer-30B0 wins(0%)(100%)5 winsMuse Spark 1.1
Benchmark scores
Grouped by capability, sorted by largest gap within each. 5 shared benchmarks.
AI Agent - Tool Usage
Muse Spark 1.1 3/3| Benchmark | Muse Glimmer-30B | Muse Spark 1.1 | Diff |
|---|---|---|---|
| Terminal-Bench 2.1 | 51.7044 / 44Thinking High (With Tools) | 8019 / 44Thinking (With Tools) | -28.30 |
| OSWorld-Verified | 65.9020 / 26Thinking High (With Tools) | 80.807 / 26Thinking (With Tools) | -14.90 |
| MCP-Atlas | 75.5020 / 38Thinking High (With Tools) | 88.101 / 38Thinking (With Tools) | -12.60 |
Coding and Software Engineer
Muse Spark 1.1 1/1| Benchmark | Muse Glimmer-30B | Muse Spark 1.1 | Diff |
|---|---|---|---|
| SWE-Bench Pro - Public | 51.2041 / 57Thinking High (With Tools) | 61.5011 / 57Thinking (With Tools) | -10.30 |
General Knowledge
Muse Spark 1.1 1/1| Benchmark | Muse Glimmer-30B | Muse Spark 1.1 | Diff |
|---|---|---|---|
| HLE | 22117 / 181Thinking High (No Tools) | 62.104 / 181Thinking (With Tools) | -40.10 |
Specs
| Field | Muse Glimmer-30B | Muse Spark 1.1 |
|---|---|---|
| Publisher | Facebook AI研究实验室 | Facebook AI研究实验室 |
| Release date | 2026-08-10 | 2026-07-09 |
| Model type | Reasoning model | Reasoning model |
| Architecture | Dense | Dense |
| Parameters | 29.6B | Not available |
| Context length | 128K+ | 1M |
| Max output | Not available | Not available |
Summary
- Muse Spark 1.1leads in:AI Agent - Tool Usage (3/3), Coding and Software Engineer (1/1), General Knowledge (1/1)
On average across the 5 shared benchmarks, Muse Spark 1.1 scores 21.24 higher.
Largest single-benchmark gap: HLE — Muse Glimmer-30B 22 vs Muse Spark 1.1 62.10 (-40.10).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.