Meta releases Muse Ember, its smallest and most cost-efficient AI model yet, saying it delivers more intelligence per dollar than Haiku 5.5 and GPT-6 Luna

Your AI assistant will cost 10 cents per million tokens to run, making complex tasks dramatically cheaper.

Meta's Superintelligence Lab has launched Muse Ember, its smallest and most cost-efficient AI model to date. Priced at 10 cents per million input tokens, Ember is designed for repetitive agent tasks such as tool calls, coding subtasks, data extraction, and document parsing. Meta claims Ember offers superior intelligence per dollar compared to competitors like Anthropic's Haiku and OpenAI's Luna, while also being cost-competitive with open-source models like DeepSeek V4.1 Flash and GLM 5.3 Flash. The model is intended to work in tandem with larger models, like Meta's Muse Wonder, in agentic systems where a larger model sets strategy and Ember executes at scale. Ember also boasts flexibility, capable of processing text, images, and audio inputs without requiring separate models for each media type. This marks the fourth release in Meta's Muse family, following Spark, Glimmer, and Wonder, and complements their successful Muse Agent consumer product.

Meta's release of Muse Ember signals a growing market emphasis on specialized, cost-efficient AI models for agentic workflows. By positioning Ember as a high-volume, low-cost execution layer, Meta suggests a future where complex AI tasks are broken down into distinct planning and execution phases, optimizing for both intelligence and affordability. This strategy directly challenges existing models by offering a clear cost advantage for repetitive yet critical agent functions, potentially influencing how other AI providers structure their offerings and pricing.

The multi-modal capabilities of Muse Ember, enabling it to process text, images, and audio without separate models, indicate a push toward more versatile and integrated AI agents. This flexibility is crucial for agents operating in real-world scenarios where data is varied and unstructured. Companies looking to deploy AI agents for tasks like customer support or data analysis will find this integration particularly appealing, as it reduces system complexity and potential points of failure, thereby accelerating the adoption of agent-based AI solutions across industries.