Meta launches Muse Ember, a low-cost model it says beats Claude Haiku and GPT-6 Luna

Meta's research lab, Meta Superintelligence Labs, launched Muse Ember on November 9, 2026, calling it the company's smallest and most efficient model to date. In the announcement, Introducing Our Smallest, Most Efficient Model: Muse Ember, Meta published benchmark comparisons claiming Muse Ember matches Claude Haiku 5.5's best scores at roughly a third of the cost and beats GPT-6 Luna across every tested setting.

What changed with Muse Ember

• Muse Ember is available now through the Meta Model API and Muse Code as model ID muse-ember-1.0, with Responses API support.
• Pricing is set at $0.10 per million input tokens, $0.004 per million cached input tokens, and $0.50 per million output tokens.
• The model supports a 1M-token input context window with up to 512K output tokens.
• Meta says Muse Ember is natively multimodal, trained on text, images, audio, and video.
• On the AutomationBench-AA benchmark, Muse Ember scores 37.8%, compared with 20.2% for GPT-6 Luna and 13.9% for Claude Haiku 5.5.
• On the OSWorld 2.1 computer-use benchmark, Muse Ember scores 69.5%, versus 68.9% for Claude Haiku 5.5 and 50.5% for GPT-6 Luna.

What it means for AI builders

Meta frames Muse Ember as a direct, numbers-backed challenge to Claude Haiku and GPT-6 Luna in the cheap, high-volume tier of agent pipelines, the layer developers rely on for tool calls, routing, and repetitive coding subtasks. By publishing head-to-head scores against both rivals, Meta gives builders who run agent harnesses at scale a new low-cost option to weigh against the models they already use for these steps.

Where Muse Ember fits in agent pipelines

Meta positions Muse Ember as the volume-handling tier meant to pair with its larger Muse Wonder model inside an agent harness. Listed use cases include tool calls, coding subtasks, routing, extraction, long-document parsing, contract review, and candidate screening, the repetitive steps that make up much of an agent's token spend.

Sources

• Introducing Our Smallest, Most Efficient Model: Muse Ember (Meta (Meta Superintelligence Labs), November 9, 2026, official)

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.