Meta launches Muse Ember for low-cost agent execution
Meta’s smallest Muse model targets repetitive agent tasks with a million-token context window and pricing designed for high-volume deployment.
What happened
Meta released Muse Ember, its smallest and most efficient model for repeated agent work such as tool calls, routing, extraction, document parsing and coding subtasks. The model is available through Meta’s Model API and Muse Code under the identifier muse-ember-1.0.
Meta lists pricing of $0.10 per million input tokens and $0.50 per million output tokens, with cached input priced at $0.004 per million tokens. The model supports a one-million-token context window and up to 512,000 output tokens, giving developers room to process large codebases or document collections in one call.
The company positions Ember as a cost-optimized complement to its larger Muse models rather than a general replacement. Meta cites Artificial Analysis results showing a 37.8% score on AutomationBench-AA at maximum reasoning, compared with 13.9% for Claude Haiku 5.5 and 20.2% for GPT-6 Luna in the same comparison. Meta also says Ember matches Haiku’s best result at roughly one-third of the cost per task.
Why it matters
The release reflects a shift in model competition toward the economics of thousands of small decisions. Agents often spend more tokens on routing, checking, formatting and calling tools than on a single difficult answer. A low-priced model that can handle those steps reliably may reduce the cost of deploying agents across products and enterprise workflows.
The claims should still be read carefully. The headline benchmarks were selected and presented by Meta, and the results depend on reasoning settings and harnesses. But the pricing is concrete and the availability is immediate. If Ember performs consistently outside Meta’s published evaluations, it could make agentic features viable in applications where premium models remain too expensive to run continuously.