# Fireworks AI launches Meta's Muse Glimmer 30B for always-on agents

> Original title: Muse Glimmer from Meta on Fireworks: Ideal for your always-on agents

- Source: Fireworks AI Blog
- Published: 2026-08-10T00:00:00.000Z
- Bad Influence: https://badinfluence.news/items/gow701f209sdsviqz181drdii
- Original: https://fireworks.ai/blog/meta-muse-glimmer-30b/

## Summary

Fireworks AI says Meta's Muse Glimmer 30B is designed for always-on agents that run many sequential tool calls and recover from failures. The 30B dense model uses sliding-window attention and two KV heads to keep the KV cache small, which the post says makes long-context agents economical at high concurrency. Meta-reported benchmarks show it ahead of Gemma 4 31B and Qwen 3.6 27B on MCP Atlas (75.5), DeepSearch QA (74.6), and SWE-Bench Pro (51.2); it is available on Fireworks serverless and on-demand, with a January 4, 2026 knowledge cutoff.

## Why it matters

The post gives architecture, serving, and benchmark details for an agent-focused model, showing the tradeoffs between tool-use performance and deployment cost.