Meta Releases Muse Glimmer, a 30B Open-Weights Agentic Model
Meta has released Muse Glimmer, a 30-billion parameter open-weights vision-language model under the permissive Apache 2.0 license. The model is specifically optimized for end-to-end agentic task completion, multi-step reasoning, and reliable tool use. This release provides the open-source community with a powerful, permissively licensed model capable of running locally on consumer hardware with 32 GB of RAM or more. It bridges the gap between proprietary models and open-weights alternatives for complex, multi-step workflows and coding tasks. Muse Glimmer demonstrates strong performance on agentic benchmarks such as DeepSearch QA, MCP-Atlas, Tau-Bench, and SWE-Bench. It also functions as a vision model capable of detailed image description and can be run locally using tools like LM Studio.
## BACKGROUND
Agentic AI models are designed to act as autonomous agents that can plan, use external tools (like APIs or databases), and execute multi-step tasks to achieve a goal. Benchmarks like MCP-Atlas and DeepSearch QA measure how well these models can handle complex, real-world workflows, such as searching the web or interacting with servers, rather than just answering simple questions.