Meta launches Muse Ember at $0.10 per million input tokens for high-volume agent subtasks

What Meta released

Meta introduced Muse Ember on 9 November 2026 and calls it its smallest and most efficient model. The company built it for the steps an agent repeats all day: tool calls, coding subtasks, routing, extraction and long document parsing. It is available now on the Meta Model API and in Muse Code under the name muse-ember-1.0, and it supports the Responses API.

The model accepts text, images, audio and video natively, so an agent can read a screenshot or work from a meeting recording without calling a separate model. It has a 1M-token context window and can return up to 512K tokens. Meta did not publish a parameter count, and the announcement does not mention open weights.

Meta positions Ember as the worker under a larger planner. Its summary is that Muse Wonder plans and Muse Ember handles the volume. Ember is also the first model Meta says it trained using Muse Wonder Full, following the specialized consumer models it described on 6 November.

Price against Meta's own lineup

Ember costs $0.10 per million input tokens, $0.004 per million cached input tokens and $0.50 per million output tokens. Prompts above 200K tokens cost three times as much, at $0.30, $0.012 and $1.50. Meta's own comparison shows why the model exists: Muse Spark 1.3 scores 37 on the Artificial Analysis Intelligence Index against Ember's 36, but costs $1.25 input and $4.25 output, and $2.39 per index task against about $0.14 for Ember.

Meta's central claim is about cost per task. It says Ember matches the best Claude Haiku 5.5 score at about a third of the cost and beats every GPT-6 Luna setting. In the chart Meta published, Ember at maximum reasoning scores 35.8 at $0.136 per task, while Haiku 5.5 at its top setting scores 36.2 at $0.411. Meta notes that the Ember costs in that chart are estimates based on its proposed pricing.

Where the benchmarks favor Ember, and where they do not

On AutomationBench-AA, a test of end-to-end business workflows run by Artificial Analysis, Meta reports 37.8 for Ember against 20.2 for GPT-6 Luna and 13.9 for Haiku 5.5. On the Coding Agent Index-AA the order is 46, 37 for Haiku and 31 for Luna. Terminal-Bench 4.0 is closer, at 40.9 for Ember and 39.2 for Haiku, with Luna at 16.4.

Ember does not lead everywhere. On FrontierCode 2.0-AA, which judges real engineering tasks on whether the code could be merged, Haiku 5.5 scores 34.9 against 27.2 for Ember. On GDP.pdf-AA, Ember and GPT-6 Luna tie at 22.8. Several other results, including Agents' Last Exam, OSWorld 2.1, PerceptionBench and the Personal Superintelligence Index, are labeled as Meta's internal evaluations, so they have not been reproduced by an outside party.

Early partner comments in the post are vendor-selected. Cloudflare's CTO says Ember ran up to 30% faster at the tail than a comparable model it tested, without naming that model. You.com reports a p95 latency of 2.2 seconds on its coding tests.

Why it matters for agent builders

Agents spend most of their tokens on small, structured steps, so the price of the cheapest capable model often sets the cost of the whole system. A team can send planning to a frontier model and route the repeated work to a model like Ember. TechCrunch notes the price also competes with open-weight options such as DeepSeek V4.1 Flash and GLM 5.3 Flash. Meta says it evaluated Ember under its Superintelligence Scaling Framework and that it stayed below the framework's high-risk thresholds. More model releases are tracked in our AI models section.

Questions people ask

How much does Muse Ember cost?

$0.10 per million input tokens, $0.004 per million cached input tokens and $0.50 per million output tokens for prompts up to 200K tokens. Longer prompts cost $0.30, $0.012 and $1.50, according to Meta.

Is Muse Ember better than Claude Haiku 5.5?

It depends on the test. Meta reports a tie at 36 on the Artificial Analysis Intelligence Index at a lower cost per task, and higher scores on AutomationBench-AA and the Coding Agent Index-AA. Haiku 5.5 scores higher on FrontierCode 2.0-AA, 34.9 against 27.2.

Are the Muse Ember weights open?

The announcement does not mention open weights. Meta offers the model through the Meta Model API and Muse Code.

Sources

1. Introducing Our Smallest, Most Efficient Model: Muse Ember (Meta, 9 Nov 2026)
2. Meta releases Muse Ember, a new lightweight model for agent execution (TechCrunch, 9 Nov 2026)