Meta (opens in new tab) on Monday released Muse Ember, which it called its "smallest and most efficient model," aimed at the repetitive work AI agents do, such as tool calls, coding subtasks, routing, data extraction and long-document parsing. The model is available through Meta (opens in new tab)'s Muse Code product and its Meta (opens in new tab) Model API.
Meta (opens in new tab) prices Ember at $0.10 per million input tokens and $0.50 per million output tokens, with a 1 million-token context window and up to 512,000 output tokens. The company says the model takes text, image, audio and video input. TechCrunch (opens in new tab) reported that the price makes Ember much cheaper than frontier rivals and cost-competitive with open-source small models.
Meta (opens in new tab) says Ember "matches Haiku's best score at about a third of the cost and beats every Luna setting," referring to Anthropic's Claude Haiku 5.5 and OpenAI's GPT-6 Luna, based on the Artificial Analysis Intelligence Index. Meta (opens in new tab)'s own charts show Ember trailing Haiku on one coding benchmark, FrontierCode 2.0. Several of the comparisons come from Meta (opens in new tab)'s internal evaluations.
Meta (opens in new tab) pitches Ember as a partner to its larger Muse Wonder model: "Muse Wonder plans, Muse Ember handles the volume," the company wrote. Meta (opens in new tab) said Ember stayed below the high-risk thresholds in its safety framework for chemical and biological, cybersecurity and loss-of-control risks. It is the fourth model under the Muse brand, after Spark, Glimmer and Wonder, according to TechCrunch (opens in new tab).