Meta’s Muse Glimmer, a 30-billion-parameter model built for multi-step agent tasks, is available on Fireworks AI through serverless and on-demand deployments. It supports text and image input and a context window of more than 128,000 tokens.
Fireworks says its small key-value cache helps reduce the cost of serving long agent sessions at scale. Meta-reported benchmark results include 74.6 on DeepSearch QA and 51.2 on SWE-Bench Pro; the post says those results used high reasoning effort. The model’s knowledge cutoff is January 4, 2026, so current information must come from tools.
Why it matters
Teams building long-running agents can try Muse Glimmer on Fireworks without arranging GPU capacity themselves, though the source does not give pricing or independent benchmark results.
