Muse Glimmer 30B

Mid-tier

Muse Glimmer 30B is Meta Superintelligence Labs' open contribution to the August 2026 wave, and it is aimed squarely at agents that run continuously on one consumer GPU. It is dense rather than mixture-of-experts, roughly 29.6 billion parameters including the vision encoder, multimodal across text and image, and released under plain Apache 2.0. At high reasoning Meta reports 76.0 on SWE-Bench Verified, 51.2 on SWE-Bench Pro, 94.7 on AIME 2026, 83.5 on GPQA Diamond, 65.9 on OSWorld-Verified, and 51.7 on Terminal-Bench 2.1, which are strong figures for a dense model at this size. Context is 131,072 tokens, notably shorter than most of its 2026 peers, so it suits long-running loops more than whole-repository passes. One practical note for anyone scripting a download: the weights live under the meta-models organization on Hugging Face, not facebook or meta-llama.

Input Price

$0.35

per 1M tokens

Output Price

$1.50

per 1M tokens

Context Window

131K

tokens

Released

2026-08

Open source

Capabilities

textvisiontool-usecodereasoning

Key Strengths

✓Apache 2.0 with no scale-based restrictions, unlike the Llama line
✓Dense 29.6B, sized for one consumer GPU
✓76.0 SWE-Bench Verified and 94.7 AIME 2026 at high reasoning
✓Multimodal text and image input
✓Built for always-on local agentic workflows

Best For

▸Always-on local agents
▸Private multimodal workflows on owned hardware
▸Desktop and computer-use automation
▸Offline or air-gapped assistants

Pricing Details

Input tokens

$0.35

per 1M tokens

Output tokens

$1.50

per 1M tokens

Estimated cost per 1K requests

$1.10

~1K input + ~500 output tokens avg

Prices are subject to change. Check the official documentation for current pricing. See the cost calculator for detailed estimates.

Open Source Model

Muse Glimmer 30B is free to download and self-host under the Apache-2.0. Hosted API pricing varies by provider (e.g., Together, Fireworks, Groq). See our open source LLM guide for deployment options.