Time Well Read
AI & technology · · 1 min

What is Meta's new Muse Ember model, and why should builders care?

Meta Platforms headquarters, Menlo Park
Image: Meta Platforms headquarters, Menlo Park, via Wikimedia Commons (see the file page for author and licence).

Meta's smallest Muse model can take in a million tokens at once and write back up to 512K. That is Muse Ember 1.0, released on 9 November 2026, though one tracker says the 10th. It costs $0.10 per million input tokens and $0.50 per million output. Cached input drops to $0.004. Meta built it for high-volume agent steps: tool calls, coding subtasks, routing, extraction, and parsing long documents. Meta frames it as the execution layer under the larger Wonder model, which does the planning. So a builder could run a lot of cheap sub-agent calls on it. It is proprietary, with no open weights. You reach it through the Meta Model API as muse-ember-1.0 or inside Muse Code. Netlify's AI Gateway added it with zero configuration. One caution. The performance claims rest on Meta's own benchmark, and no independent agentic evaluation had been reported yet.

Sources: ai-tldr.dev, netlify.com

Saved something worth reading? Add it

More like this