Meta

AI · Enterprise Software

Launch

Meta launches Muse Ember, a small AI model built to undercut Haiku, Luna

Pricing it to match cost-per-task leaders signals Meta is now competing on efficiency, not just raw intelligence scores.

• Meta released Muse Ember on November 9, 2026, its smallest and most efficient model yet, available immediately via Muse Code and the Meta Model API.
• Ember is built for the repetitive work AI agents handle constantly: tool calls, coding subtasks, routing, data extraction, and long-document parsing.
• Pricing is $0.10 per million input tokens and $0.50 per million output tokens, matching list prices for Claude Haiku 5.5 and GPT-6 Luna, with cached input at $0.004 per million.
• On the Artificial Analysis Intelligence Index, Ember matches Haiku 5.5's top score of 36 at about a third of the cost per task ($0.14 vs $0.41), and beats Luna's 28.
• Ember trails Haiku on one coding benchmark, FrontierCode 2.0, scoring 27.2% versus Haiku's 34.9%, showing the efficiency gains aren't universal across tasks.
• Ember is the fourth model in Meta's Muse family, following Spark in April, open-source Glimmer in August, and frontier model Wonder in October, with Wonder meant to plan while Ember executes at scale.
• As AI labs race to build agents that run thousands of cheap, repetitive calls, winning on cost-per-task efficiency rather than peak intelligence is becoming its own competitive battleground.