Meta ships Muse Ember at Claude Haiku 5.5's $0.10 / $0.50, and Artificial Analysis scores the two level at 36

Meta released Muse Ember today, its smallest and most efficient model, built for the work agents repeat all day: tool calls, coding subtasks, routing, extraction and long documents. It is on Meta Model API now as muse-ember-1.0, and in Muse Code. Mark Zuckerberg announced it on X at 17:54 UTC; @AIatMeta followed at 18:04.

THE PRICE AND SPECS

• $0.10 input / $0.50 output per 1M tokens, $0.004 for cached input. That is the same sticker as Claude Haiku 5.5 (for prompts up to 100K tokens).
• 1M-token context window and up to 512K output tokens, per Meta.
• Text, image, audio and video in; text out. It supports the Responses API.
• Meta says Ember is the first model trained using Muse Wonder Full, the stronger Wonder variant in Muse Code.
• OpenCode added it to its Zen gateway the same day, at the same price.

THE INDEPENDENT NUMBER

Artificial Analysis scores Muse Ember (Max) 36 on its Intelligence Index, the same as Claude Haiku 5.5 (Max); its model pages list Haiku 5.5 first and Ember second of 32 on intelligence (read on its site today). It puts Ember's cost at $0.14 per index task. Meta's "about a third of the cost" is per task, not per token: the two list at the same price, and Meta's chart, built from Artificial Analysis results, has Haiku 5.5 at $0.41 a task at max effort.

META'S OWN BENCHMARKS

These are Meta's figures, at max reasoning effort, against Claude Haiku 5.5 and GPT-6 Luna:

• Terminal-Bench 4.0: 40.9, against 39.2 and 16.4.
• AutomationBench-AA: 37.8, against 13.9 and 20.2.
• OSWorld 2.1 (partial credit): 69.5, against 68.9 and 50.5.
• Meta's internal Personal Superintelligence Index: 46.8, against 36.1 and 37.0.

SAFETY

Meta says it evaluated Ember under its Meta Superintelligence Scaling Framework for chemical and biological, cybersecurity and loss-of-control risks, and that it stayed below the framework's high-risk thresholds in every category assessed.

HOW WE CAUGHT IT

Our radar saw muse-ember-1.0 on the public copy of Meta's own model list at 17:54:06 UTC, 34 seconds before Zuckerberg's post, and sent the alert at 17:55:06. It is now a row on /lead-time.

WHAT WE DO NOT KNOW

• How it scores on other public boards. Only Artificial Analysis has a score so far.
• Whether Meta will release Ember's weights. Its post does not say.