Meta introduces Muse Ember to power cost-effective AI agents

Meta Platforms Inc. today introduced Muse Ember, an entry-level artificial intelligence model that costs significantly less than the competition.

Notably, it is the first AI that the company has trained with Muse Wonder. The latter algorithm is the flagship of Facebook’s parent company’s large language models. Meta didn’t specify what part of the workflow it automated with Muse Wonder.

One possibility is that the company used it to perform post-training, the task of improving a model’s capabilities by training it on additional data. Thinking Machines Labs Inc. recently demonstrated that even a relatively small LLM can run post-training runs consistently. Anthropic PBC again last week revealed that its AI models could soon gain the ability to significantly accelerate machine learning research.

Applications often use frontier LLMs like Muse Wonder as part of AI ensembles. When a user enters a query, Frontier LLM breaks it down into smaller tasks and assigns them to agents based on a more cost-effective model. According to Meta, Muse Ember is specifically optimized for the latter role.

The company evaluated the model using the Artificial Analysis Index, which aggregates nine AI benchmarks from use cases such as coding and financial analysis. Meta says Muse Ember rivaled Anthropic’s entry-level Claude Haiku 5.5 at about a third of the price. It also outperformed OpenAI Group PBC’s comparable GPT-6 Luna algorithm.

During the same internal evaluation, Meta Muse tested Ember and the two competing models in nine other benchmarks. All three algorithms were in “Max” mode, meaning they used the maximum amount of available hardware rather than prioritizing cost effectiveness. Muse Ember outperformed Haiku and Luna in seven benchmarks.

The area where the model appears to have the greatest advantage is business task automation. While most of Muse Ember’s benchmark results were only a few percentage points better than Haiku and Luna, it achieved a 17% lead on AutomationBench-AA. It’s a collection of 657 tasks spanning popular cloud apps like Gmail and Salesforce.

Muse Ember has a context window of 1 million tokens, allowing the ingestion of large business documents. Users can also insert images, audio, and video into their prompts. The model generates responses up to 512,000 tokens long.

Muse Ember is available via Meta’s Model API for 10 cents per million input tokens and 50 cents per million output tokens. A built-in cache can reuse frequently used prompts to avoid duplicate processing. When the feature is active, the price of Muse Ember drops to 0.4 cents per million input tokens.

Image: Meta