META'S SMALLEST AI MATCHES CLAUDE HAIKU AT A THIRD OF THE COST

Meta released Muse Ember on November 9, its smallest and most efficient model — muse-ember-1.0 — built for the repetitive work agents do all day: tool calls, coding subtasks, routing, extraction and long-document parsing. It takes a 1 million-token context, generates up to 512,000 tokens, and costs $0.10 per million input tokens and $0.50 per million output.

Meta says Ember matches Claude Haiku 5.5's best Artificial Analysis Intelligence Index score of 36 at about a third of the cost per task, and beats OpenAI's GPT-6 Luna at every reasoning setting Meta plotted — Meta's own comparisons, pending independent validation. It is the first model trained using Muse Wonder Full, and it is available now on the Meta Model API and Muse Code.