Meta introduces Muse Ember, a cost-efficient model for agentic tasks

Meta releases Muse Ember, its smallest and most efficient model, designed for complex reasoning execution and repetitive agent tasks such as tool calls, coding subtasks, and document parsing. It is the first model trained using Muse Wonder Full.
Muse Ember matches the best score of Claude Haiku 5.5 at approximately one-third of the cost and outperforms every reasoning setting of GPT-6 Luna on the Artificial Analysis Intelligence Index.
The model achieves a score of 33.9 on the Artificial Analysis Intelligence Index at a cost of $0.116 per task, compared to Claude Haiku 5.5's score of 34.3 at $0.248 per task.
Muse Ember leads in agentic benchmarks, scoring 37.8% on AutomationBench-AA and 61.6% on Agents' Last Exam, surpassing GPT-6 Luna and Claude Haiku 5.5.

Performance and Capabilities

The model features a 1M-token context window and supports up to 512K output tokens, enabling it to process full codebases or document sets in a single call.
It natively handles text, images, audio, and video, achieving 80.6% on PerceptionBench and 69.5% on OSWorld 2.1 for computer use tasks.
In coding evaluations, Muse Ember scores 40.9% on Terminal-Bench 4.0 and 46 on Coding Agent Index-AA, though it trails Claude Haiku 5.5 on FrontierCode 2.0-AA (27.2% vs 34.9%).

Available immediately on the Meta Model API and Muse Code as muse-ember-1.0.
Pricing is $0.10 per million input tokens, $0.004 per million cached input tokens, and $0.50 per million output tokens.
The model is also available on Google Cloud and Oracle Cloud. Partners including Box, Cline, Oracle, Cloudflare, and You.com reported early testing results.