Meta has launched Muse Ember, a low-cost AI model priced at $0.10 per million input tokens. The new Meta AI model targets developers and businesses looking for affordable solutions to run AI agents. Its launch also adds to the competition with OpenAI and Anthropic in the AI model market.
Announced on November 9, 2026, Muse Ember is built to handle repetitive tasks in AI agent workflows. These include coding subtasks, data extraction, document processing, and tool calls. Meta aims to help developers manage large volumes of AI tasks while keeping inference costs low.
Muse Ember is part of Meta’s Muse AI model family. Developers can access it through the Meta Model API and Muse Code. The model also offers a context window of up to 1,048,576 tokens and accepts text, images, audio, and video as inputs.
With Muse Ember, Meta is targeting businesses that need affordable AI models for large-scale automation. The company says the model delivers competitive benchmark results while costing less to run than several larger AI models.
Meta Sets Muse Ember AI Pricing at $0.10 Per Million Input Tokens
Meta has introduced competitive pricing for its Muse Ember AI model. The company charges $0.10 per million input tokens, $0.50 per million output tokens, and $0.004 per million cached input tokens.
These rates apply to requests with up to 200,000 input tokens. For longer requests, Meta charges $0.30 per million input tokens and $1.50 per million output tokens. The cached input rate increases to $0.012 per million tokens.
The low pricing makes Meta Muse Ember an affordable option for developers handling large volumes of AI requests. It could also help businesses reduce the cost of running AI agents that perform repetitive tasks.
According to Meta, Muse Ember costs around $0.14 per task in Artificial Analysis evaluations. Its larger Muse Spark 1.3 model costs approximately $2.39 per task.
Meta is targeting developers and businesses that need AI agents to handle multiple small tasks at scale. The pricing could make Muse Ember a practical choice for applications that require frequent AI operations.
Meta Muse Ember AI Model Is Built for AI Agents and Repetitive Tasks
Meta Muse Ember is designed to handle smaller tasks within larger AI agent workflows. It focuses on tasks that require frequent processing rather than complex reasoning.
The model can make tool calls, complete coding tasks, extract information, route requests, and process long documents. It can also analyze large collections of images, files, and records.
Muse Ember has a context window of up to 1,048,576 tokens and supports up to 524,288 output tokens. This allows developers to process large amounts of data in a single request.
The AI model accepts text, images, audio, video, and PDF files as input. However, it generates text-only responses.
Meta expects developers to use Muse Ember alongside larger models such as Muse Wonder and Muse Spark. These models can handle complex reasoning and planning, while Muse Ember manages smaller tasks across multiple operations.
For example, an AI agent could use a larger model to plan a customer support workflow. Muse Ember could then process individual customer queries, extract relevant details, and route requests to the right teams. This approach could help businesses run AI agents at scale while keeping operating costs low.
Meta Muse Ember Outperforms GPT-6 Luna and Claude Haiku 5.5 in Several AI Benchmarks
Meta claims its Muse Ember AI model can compete with rival models from Anthropic and OpenAI across several AI benchmarks. The company compared its performance with Anthropic’s Claude Haiku 5.5 and OpenAI’s GPT-6 Luna.
According to benchmark results released with the announcement, Muse Ember scored 36 on the Artificial Analysis Intelligence Index. Claude Haiku 5.5 also scored 36, while GPT-6 Luna recorded 28.
Muse Ember also delivered higher scores in several AI agent and coding tests. In AutomationBench-AA, which measures how well AI models complete business workflows, Muse Ember scored 37.8%. This was higher than GPT-6 Luna’s 20.2% and Claude Haiku 5.5’s 13.9%.
The model also scored 40.9% on Terminal-Bench 4.0. Claude Haiku 5.5 followed with 39.2%, while GPT-6 Luna scored 16.4%. However, Muse Ember did not lead every benchmark. Claude Haiku 5.5 performed better in FrontierCode 2.0-AA, scoring 34.9% compared with Muse Ember’s 27.2%.
The results show that Meta Muse Ember can compete with established AI models in several coding and AI agent evaluations. However, its performance varies depending on the benchmark and the type of task.
Meta Makes Muse Ember Available Through Its Model API and Muse Code
Meta has released Muse Ember through its Meta Model API and Muse Code. Developers can access the model using the model ID muse-ember-1.0.
The model also supports the Responses API and is available through Google Cloud and Oracle Cloud. This allows developers to integrate Muse Ember into their applications and AI agent workflows.
Meta has also added Muse Ember to its Muse Code subscription plans. Developers can use the model for coding tasks, automation, and other AI-based operations.
According to Meta, Muse Ember is the first publicly released model trained using Muse Wonder Full. Its release is part of the company’s wider effort to develop smaller and more efficient AI models.
Meta also assessed Muse Ember for potential cybersecurity, chemical, and biological risks. The company evaluated possible loss-of-control risks as well. It said the model remained below its high-risk thresholds across all assessed categories.
Meta Muse Ember Puts Pressure on OpenAI and Anthropic With Cheaper AI Pricing
The launch of Muse Ember adds to the growing competition between Meta, OpenAI, and Anthropic in the AI market. Major AI companies are competing to build more capable models while reducing the cost of running them.
This is important for businesses using AI agents, as a single workflow can require multiple model requests. These repeated requests can increase AI operating costs, especially when companies handle thousands of tasks daily.
Meta aims to address this challenge with Muse Ember’s low token pricing and focus on repetitive tasks. Its benchmark results also show that smaller AI models can deliver competitive performance in specific coding and automation tasks.
However, Muse Ember is not designed to replace larger AI models. Meta expects developers to use it alongside more advanced systems that handle complex reasoning and planning.
With Muse Ember, Meta is expanding its AI model lineup and competing on both performance and pricing. Its long-term success will depend on how well the model performs in real-world applications and whether businesses adopt it for large-scale AI agent workflows.
