1min.

1MIN · AI & AGENTS

Muse Ember targets repeatable agent work at lower token prices

Meta’s Muse Ember is a hosted model aimed at repeated agent tasks such as extraction, routing and coding subtasks. Prompts up to 200,000 tokens cost $0.10 per million input tokens and $0.50 per million output tokens; longer prompts cost three times as much. It offers a million-token context window, but low token prices alone do not establish reliable task completion.

Meta AI · Nov 9, 2026

Why it matters

High-volume agent steps deserve their own quality-and-cost evaluation instead of automatically using the largest model.

More context

Available through Muse Code and Meta Model API as muse-ember-1.0. Try a small set of real routing or extraction cases and count retries, failures and total tokens. Meta’s evaluations use different agent harnesses for some coding comparisons.

Keep in mind

Released hosted model with paid API usage. Pricing depends on prompt length; vendor benchmark rankings are not independent evidence for your workflow, and the announcement does not provide downloadable weights.

Original read

More engineering ideas, one minute at a time.

Explore more stories