Meta releases Muse Ember, its smallest model, at $0.10 per million input tokens, and claims more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna

Meta released Muse Ember on 9 November and calls it its smallest and most efficient model. It is aimed at the repetitive work agents do, such as tool calls, coding subtasks, routing, extraction and reading long documents. It is available as muse-ember-1.0 on Muse Code and the Meta Model API, and Meta says the Muse models are now on Google Cloud and Oracle Cloud.

The context window is 1 million tokens, with up to 512,000 tokens of output, and it takes text, images, audio and video. It costs $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 200,000 tokens, and $0.30 and $1.50 above that.

Meta's blog says Muse Ember “delivers more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna at any reasoning setting”. That is a claim about cost, not raw ability: Meta's model page lists Muse Ember at 27.2 on FrontierCode 2.0-AA, a test of real engineering tasks, and 36 on the Artificial Analysis Index. The Decoder reports Claude Haiku 5.5 scores 34.9% on the same coding test.

Muse Ember follows Muse Wonder, Meta's first frontier model, which the Switchboard ran on 13 October, and adds a cheap tier for companies running many agent steps.

• Confirmed Released 9 November as muse-ember-1.0, with a 1M-token context and $0.10/$0.50 per million tokens up to 200K tokens ($0.30/$1.50 above). Meta
• Claimed Meta says Muse Ember delivers more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna at any reasoning setting. Meta
• Confirmed Meta's model page lists 27.2 on FrontierCode 2.0-AA and 36 on the Artificial Analysis Index. Meta model page
• Reported Claude Haiku 5.5 scores 34.9% on FrontierCode 2.0-AA against Muse Ember's 27.2%, The Decoder reports. The Decoder