Meta released Muse Ember on November 9, 2026, for repeatable work inside agent workflows. It's available on Meta Model API and in Muse Code. If your agent already runs on Muse Spark 1.3, try Ember on one worker route before changing the main loop.
Compared with Muse Spark 1.3
Route work with a clear input and a checkable result to Ember: classify support tickets, extract fields from PDFs, make repeated tool calls, or run a focused coding subtask. Keep Spark for planning or knowledge-heavy turns, and consider Wonder for difficult multi-file changes. Ember accepts text, image, video, audio, and PDF inputs and returns text, with a 1,048,576-token context window. See Meta's model guide.
API Migration
In an existing Meta Model API request, set model to muse-ember-1.0. Keep your Responses, Chat Completions, or Messages endpoint and request shape. For a fair comparison, set a reasoning effort both models support. Spark 1.3 defaults to high, while Ember defaults to max. For a side-by-side test against Spark 1.3 Contributor, try high on both models. Ember is Standard only on Model API, so review token rates and data use if you're switching from a Spark Contributor model. See Meta's switching guide and pricing modes.
Long Context Pricing
The Meta Model API rates below are USD per million tokens:
| Tokens | Up to 200,000 input tokens | Over 200,000 input tokens |
|---|---|---|
| Uncached input | $0.10 | $0.30 |
| Cached input | $0.004 | $0.012 |
| Output | $0.50 | $1.50 |
Cached input and stored conversation history carried forward via previous_response_id count toward the threshold. Once the request crosses it, the higher rates apply to the whole request. For example, 190,000 uncached input tokens and 10,000 billed output tokens cost $0.024 in token charges; 250,000 uncached input tokens and the same output cost $0.09. Pass only the context each worker needs and measure cost per successful task before routing more traffic.
