Muse Glimmer-30B
Meta · open weight · muse-glimmer-30b
Muse Glimmer-30B is Meta Superintelligence Labs' open-weight, dense multimodal model for autonomous agentic work on consumer hardware. Its 29.6-billion-parameter causal transformer includes a roughly 1.8B-parameter ViT-G/14 perception encoder, accepts interleaved text and images, produces text, and supports a 131,072-token context window. It combines controllable multi-step reasoning, schema-based tool use, coding, failure recovery, and compatibility with agent scaffolds such as OpenClaw and Hermes Agent. Meta also releases 4-bit variants sized for 24 GB and 32 GB systems and a DFlash speculative-decoding drafter. The model is trained on more than 100 languages and released under Apache 2.0.
Benchmark scores
| Benchmark | Score |
|---|---|
| AIME 2026 | 0.95 |
| Siren AgentDojo Utility | 0.94 |
| GPQA | 0.83 |
| SWE-Bench Verified | 0.76 |
| MCP Atlas | 0.76 |
| DeepSearchQA | 0.75 |
| MMMU-Pro | 0.74 |
| Humanity's Last Exam | 0.22 |
Pricing
- DeepInfra$0.30 / $1.20
Input / output per 1M tokens
AA metrics
No Artificial Analysis link yet.
Arena Elo
- No Arena snapshot linked.
