# Muse Ember: Meta's smallest model for agent work, at $0.10 per million input tokens

Canonical: https://theagenticweekly.com/news/meta-muse-ember · Evidence: Reported · First seen: 2026-11-09 18:01 UTC · Source published: 2026-11-09 18:00 UTC · Brief published: 2026-11-09 19:01 UTC · Updated: 2026-11-09 19:01 UTC · Author: Agentic Weekly

> Muse Ember is Meta's smallest and cheapest Muse model, released Nov 9, 2026 as muse-ember-1.0 on the Meta Model API and in Muse Code. It costs $0.10 per million input tokens and $0.50 per million output tokens, has a 1M-token context window, and Meta pitches it as the high-volume executor paired with Muse Wonder or Muse Spark.

Models in this brief: [Muse Ember](https://theagenticweekly.com/models/muse-ember), [Muse Wonder](https://theagenticweekly.com/models/muse-wonder), [Muse Wonder Full](https://theagenticweekly.com/models/muse-wonder-full), [Claude Haiku 5.5](https://theagenticweekly.com/models/claude-haiku-5-5), [GPT-6 Luna](https://theagenticweekly.com/models/gpt-6-luna)

## What changed

- Meta released Muse Ember as muse-ember-1.0 on the Meta Model API and in Muse Code, priced at $0.10 per million input tokens, $0.004 per million cached input tokens and $0.50 per million output tokens.
- It has a 1,048,576-token context window and up to 524,288 output tokens, takes text, image, video, audio and PDF input, and supports the Responses, Chat Completions and Messages endpoints.
- Meta says Ember matches Claude Haiku 5.5's best Artificial Analysis score at about a third of the cost; its own chart puts Ember at 35.8 for $0.136 per task against Haiku 5.5 at 36.2 for $0.411, with Ember costs marked as estimates.
- In Meta's table at max reasoning, Ember scores 37.8 on AutomationBench-AA versus 13.9 for Haiku 5.5 and 20.2 for GPT-6 Luna, but trails Haiku on FrontierCode 2.0-AA (27.2 versus 34.9).
- The docs say Ember accepts the same request body as Muse Spark, so switching is mostly a model ID change; it runs in Standard mode only, and requests over 200K input tokens are billed at higher long-context rates.
- Meta says Ember is the first model trained using Muse Wonder Full and that its safety evaluations found it below the high-risk thresholds of its scaling framework.

## Who it matters for

Teams running agent harnesses with many small steps: sub-agent coding tasks, tool calls, request routing, and extraction over large batches of documents or images. Meta positions it as the cheap executor under a planner such as Muse Wonder or Muse Spark, so existing Muse Spark users can try it by changing the model ID.

If you mainly need hard single-shot coding, Meta's own table shows Claude Haiku 5.5 ahead on FrontierCode 2.0-AA, and every benchmark here is reported by Meta, not independently run.

## In the source's words

> "it matches Haiku's best score at about a third of the cost and beats every Luna setting."
> (Meta, Introducing Muse Ember, https://research.meta.ai/blog/muse-ember)

> "Muse Ember accepts the same request body as Muse Spark, so switching most examples is a model ID change."
> (Meta Model API docs, Muse Ember, https://dev.meta.ai/docs/muse-ember)

## What we have not verified

- We did not run this release ourselves.
- All benchmark scores and cost-per-task figures are Meta's own reporting of Artificial Analysis and internal evaluations; the blog notes Ember costs in its chart are estimates at proposed pricing.
- We did not check the long-context price tier for requests over 200K input tokens or the rate limits.
- Partner quotes in the announcement (Box, Cline, Oracle, Cloudflare, You.com) were not checked with those companies.

## Timeline

- Source published: 2026-11-09 18:00 UTC
- First seen by our watcher: 2026-11-09 18:01 UTC (via TechCrunch: AI)
- Brief published: 2026-11-09 19:01 UTC

## Sources

- [Meta: Introducing Muse Ember](https://research.meta.ai/blog/muse-ember) (primary source) (fetched 2026-11-09)
- [Meta Model API docs: Muse Ember](https://dev.meta.ai/docs/muse-ember) (primary source) (fetched 2026-11-09)
- [TechCrunch: Meta releases Muse Ember](https://techcrunch.com/2026/11/09/meta-releases-muse-ember-a-new-lightweight-model-for-agent-execution/) (fetched 2026-11-09)

## Related

- [Muse Ember: specs, price and alternatives](https://theagenticweekly.com/models/muse-ember)
- [Muse Wonder: specs, price and alternatives](https://theagenticweekly.com/models/muse-wonder)
- [Muse Wonder Full: specs, price and alternatives](https://theagenticweekly.com/models/muse-wonder-full)
- [Claude Haiku 5.5: specs, price and alternatives](https://theagenticweekly.com/models/claude-haiku-5-5)
- [GPT-6 Luna: specs, price and alternatives](https://theagenticweekly.com/models/gpt-6-luna)
- [All news briefs](https://theagenticweekly.com/news)
- [This week's releases](https://theagenticweekly.com/this-week)
- [New AI models this week](https://theagenticweekly.com/models/this-week)
- [AI model releases by month](https://theagenticweekly.com/models/releases)
- [Best LLM for coding](https://theagenticweekly.com/best/llm-for-coding)

## How to cite

Agentic Weekly, "Muse Ember: Meta's smallest model for agent work, at $0.10 per million input tokens", published 2026-11-09 19:01 UTC, https://theagenticweekly.com/news/meta-muse-ember