Meta Launches Muse Ember AI Model for Coding and Automation Tasks

Meta has introduced Muse Ember, a new AI model designed for developers and enterprises to help automate routine tasks using AI agents. The model is specifically built to handle smaller, discrete tasks within larger workflows—such as reading files, running code tests, extracting information, and invoking external tools. According to Meta, Muse Ember delivers performance comparable to larger models but at a lower cost, making it highly suitable for repetitive tasks performed by AI agents. The model is now available via Meta’s developer platform.

Meta Muse Ember: Capabilities and Benchmarks
In a blog post, Meta explained that Muse Ember is tailored for the specific, smaller-scale tasks involved in operating AI agents, including reading documents, modifying code, executing tests, and retrieving information. The company added that the model can also utilize external tools and manage day-to-day business operations—for instance, looking up specific details in corporate records and updating that information across various applications.

This model supports a context window of 1 million tokens and can generate up to 512,000 output tokens. It accepts text, images, audio, and video as input, enabling it to process various types of content. Developers can use it to analyze long-form documents, examine screenshots, and work with large codebases—all without needing to break them down into multiple smaller requests.

Meta has released benchmark results for Muse Ember, comparing it against Anthropic’s Claude Haiku 5.5 and OpenAI’s GPT-6 Luna. The model achieved a score of 36 on the AI ​​Analysis Index (AII)—tying with Claude Haiku 5.5 and outperforming GPT-6 Luna, which scored 28. Meta also claims that while Muse Ember’s performance is on par with Claude Haiku 5.5, it costs only about one-third as much to run.

In the AutomationBench-AA test, designed to evaluate an AI model’s ability to handle automated tasks, Muse Ember scored 37.8%—surpassing both GPT-6 Luna (20.2%) and Claude Haiku 5.5 (13.9%). In the Terminal-Bench 4.0 test, Muse Ember achieved a score of 40.9%, slightly ahead of Claude Haiku 5.5’s 39.2%. Conversely, Claude Haiku 5.5 performed better on the FrontierCode 2.0-AA test, scoring 34.9% compared to Muse Ember’s 27.2%.

Meta stated that Muse Ember is the first model it has trained using “Muse Wonder Full.” The company also conducted safety evaluations covering cybersecurity threats, harmful instructions, and other potential risks, reporting that the model did not reach any high-risk thresholds.

Availability of Meta Muse Ember
Meta has released Muse Ember via its Model API and Muse Code, platforms through which developers can access the muse-ember-1.0 model. The company charges $0.10 (approximately 9.36 rupees) per million input tokens and $0.50 (approximately 47 rupees) per million output tokens. For cached input tokens, the cost is $0.004 (approximately 0.37 rupees) per million tokens. Meta has also added support for a response API and has deployed the model on Google Cloud and Oracle Cloud.

Leave a Comment