Muse-Glimmer-30B
#49 in Open-Source-SpraakmodelleUnknown · 30b · siet 10. August 2026 · 2× · tolest 27. Okt. 2026
22
Momentum
Muse Glimmer 30B is an open-weight, dense multimodal language model (text and image input, text output) from Meta Superintelligence Labs, designed for agentic tasks on consumer hardware. It has about 29.6B parameters including a roughly 1.8B vision encoder and a 131,072-token context window. The weights are released under the Apache 2.0 license on Hugging Face.
Momentum-Verloop
30.07.28.10.
Features
| Key Benchmark (%) | SWE-Bench Verified 76.0%; AIME 2026 94.7%; GPQA Diamond 83.5% (high reasoning, per Meta) |
| Context Window (Tokens) | 131,072 (131K) |
| License | Apache 2.0 |
| Multimodality | Input: text and image (video as sampled frames, no audio); output: text |
| Platform | Hugging Face (BF16, GGUF, FP8/NVFP4); runs with vLLM and llama.cpp; API via third parties (e.g., DeepInfra, OpenRouter) |
| Price per 1M Tokens | Third-party list price: $0.30 input / $1.20 output (DeepInfra, OpenRouter); no token fee when self-hosted |
| Release Date | August 10, 2026 |