Meta launches Muse Ember at one-third the cost of rivals

Meta has launched Muse Ember, its smallest and most efficient AI model yet. Built for complex reasoning and the repetitive daily work of AI agents, it is now live on Meta Model API, Google Cloud, and Oracle Cloud. On the Artificial Analysis leaderboard, Muse Ember matches Claude Haiku 5.5 at just one-third of the cost and beats GPT-6 Luna in many tests. It has a 1 million token context window and can process a full codebase or a set of documents in one go. The model natively understands text, images, audio, and video, so agents can read screenshots, scanned documents, and meeting recordings without other models. It can also run hundreds of coding subtasks in one session. Safety tests show its risk level stays below the high-risk threshold. Pricing is $0.10 per million input tokens and $0.50 for output.