Meta Glimmer Reasoning Report - SWE-Bench Verified: leaderboard
Meta Muse Glimmer technical report,October19,AppendixTable5: separate low/mid/high/xhigh reasoning runs; AppendixA defines each metric and source. Only reported cells are emitted, missing settings stay missing. τ³Banking,GDPval-AA,SciCode,TerminalBench,GPQA,HLE,MMMU-Pro andAA-LCR include externally sourced evaluations rather than a single internal harness.
Metric: Published Score (%). Source: ai.meta.com. Saturation forecast: Around January 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Muse Glimmer 30B (High) | 76 | #179 (Muse Glimmer 30B) |
| 2 | Muse Glimmer 30B (xHigh) | 75.3 | #179 (Muse Glimmer 30B) |
| 3 | Muse Glimmer 30B (Medium) | 75.2 | #179 (Muse Glimmer 30B) |
| 4 | Muse Glimmer 30B (Low) | 74.6 | #179 (Muse Glimmer 30B) |
No result here: #4 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=meta-glimmer-reasoning-report-swe-bench-verified · How It Works · Data refreshed daily, snapshot 2026-10-22.