Muse Ember 1.0

Meta introduced Muse Ember on Nov. 9, 2026 as "our smallest and most efficient model," the first model trained using Muse Wonder Full, built for "the work agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing." It has a 1M-token context window, up to 512K output tokens, and Meta says it natively understands text, images, audio, and video. Muse Ember is available now on Muse Code and the Meta Model API as muse-ember-1.0, alongside Muse Spark and Muse Wonder on Google Cloud and Oracle Cloud. Meta reports an Artificial Analysis Intelligence Index score of 36, tied with Claude Haiku 5.5 and ahead of GPT-6 Luna at 28, with an estimated cost of about $0.14 per task versus $2.39 for Muse Spark 1.3.

Meta introduced Muse Ember on November 9, 2026 as its smallest and most efficient model, aimed at repeated agent work such as tool calls, coding subtasks, routing, extraction, and long document parsing. Meta says it delivers more intelligence per dollar than Claude Haiku 5.5 and GPT-6 Luna at any reasoning setting, matching Haiku's best score at roughly a third of the cost. Using Artificial Analysis data, Meta reports Muse Ember at about $0.116 per task with a 33.9 score and $0.136 per task at 35.8, versus Claude Haiku 5.5 reaching 36.2 at $0.411 and GPT-6 Luna topping out at 27.8. Meta claims Ember sits on the efficiency frontier, with no model scoring higher for less per task, and the figures are estimates at proposed pricing.

Fello AI's review of Meta Muse Ember, released November 9, 2026, says the model is not cheaper per token than rivals, listing at $0.10 input and $0.50 output per million tokens like Claude Haiku 5.5 and GPT-6 Luna. Its advantage comes from generating fewer tokens: 115M output tokens on the Artificial Analysis run versus Haiku's 298M. The review highlights that Ember wins seven of Meta's ten benchmark rows, ties two, and loses one, with Claude Haiku 5.5 ahead on FrontierCode 2.0 at 34.9 versus 27.2. It also flags that prompts beyond 200,000 tokens triple the price to $0.30 input and $1.50 output per million, a detail Meta documents on its developer page rather than the launch post.

Simon Willison's independent analysis of the Muse Ember launch names the version Muse Ember 1.0 and lays out its tiered pricing: up to 200k input tokens the model costs $0.10 per million input tokens, $0.004 cached input, and $0.50 output, while longer prompts jump to $0.30/$0.012/$1.50. Context window reaches one million input tokens. He notes Ember sits exactly on the Artificial Analysis Pareto line of intelligence versus cost per task, and that it exposes six reasoning levels from minimal through max, with a high setting failing to return a complete SVG in his test. He contrasts it with same-day Claude Fable 5.2 pricing of $8 and $40 per million tokens.

Meta unveiled Muse Ember as its smallest addition to its model family, a lightweight workhorse for agentic applications, released as muse-ember-1.0 through Muse Code and the Meta Model API at $0.10 per million input tokens, $0.004 cached input, and $0.50 per million output, with a 1-million-token context window and up to 512K output tokens. Meta puts its cost near $0.14 per Artificial Analysis benchmark task. Partner testing highlighted latency and efficiency: Cloudflare CTO Dane Knecht reported Muse Ember completed every coding test at a p95 of 2.2 seconds, roughly 4x to 13x faster than leading open-weight models, while using 36-48% fewer output tokens. Meta says it is purpose-built for repetitive subagent work, and that it was vetted under the Meta Superintelligence Scaling Framework with risk evaluations below high-risk thresholds.

Simon Willison notes that Meta's Muse Ember entry into the Claude Haiku and GPT-6 Luna category is priced at $0.10 per million input tokens and $0.50 per million output for short prompts, with six reasoning levels from minimal through max. He observes that all three models support up to a million input tokens but raise prices past a threshold, and that Ember sits between GPT-6 Luna and Haiku 5.5 once that threshold is accounted for. He also notes that Ember's price rises above 200k prompts to $0.30 input, $0.012 cached, and $1.50 output, and that Meta calls out its position on the Artificial Analysis efficiency frontier, where no model scores higher for less per task. The same day, Anthropic shipped Claude Fable 5.2 at $8 and $40 per million, roughly 80 times higher.

Meta Superintelligence Labs released Muse Ember, its lightest and cheapest Muse model, as the first Meta model distilled from Muse Wonder Full and designed as a subagent paired with Wonder. It is served as muse-ember-1.0 through the Meta Model API and Muse Code, supports the Responses API, and carries a 1 million-token context window with native text, image, audio, and video input. Pricing is $0.10 per million standard input tokens, $0.004 cached, and $0.50 output, which Artificial Analysis says is roughly 60% below Haiku 5.5's cached rate; combined with about one-third as many output tokens, that yields an estimated $0.14 per index task versus $0.41 for Haiku 5.5. Ember scores 36 on the Artificial Analysis Intelligence Index, matching Haiku 5.5, but reports a weakness of 21.1% on AA-Omniscience, declining 64% of knowledge questions to avoid hallucinations.