Meta has unveiled Muse Ember, its smallest and most cost-efficient AI model yet, engineered to excel at routine tasks such as tool calls, coding subtasks, request routing, data extraction, and parsing large documents. According to independent benchmarks, Ember matches the intelligence level of Anthropic’s Claude Haiku 5.5 while cutting the cost per task to about a third.
- Muse Ember matches Claude Haiku’s intelligence score at one-third the cost per task
- Designed for high-volume, token-heavy tasks like coding and data extraction
- First Meta model trained using the larger Muse Wonder Full and supports multimodal inputs
What happened
Meta has launched Muse Ember, its smallest and cheapest AI model to date, targeting the repetitive tasks AI agents handle daily, such as tool integration, routing, coding subtasks, and parsing long documents. Ember is priced competitively at $0.10 per million input tokens and $0.50 per million output tokens, placing it alongside Anthropic’s Claude Haiku 5.5 and OpenAI’s GPT-6 Luna on cost.
Independent benchmarking from Artificial Analysis found Muse Ember achieves an intelligence index score of 36 on its highest setting, equal to Claude Haiku 5.5 and better than Luna’s 28. At about $0.14 per task, Ember maintains roughly a third of the cost per task compared to Haiku’s $0.41, making it one of the most cost-efficient models available for high-volume processing.
Why it matters
Muse Ember’s combination of cost efficiency and strong performance makes it a compelling choice for applications requiring frequent, token-heavy operations. Its ability to handle text, images, audio, and video input while processing up to one million tokens simultaneously opens opportunities for complex workflow automation and large-scale document analysis.
Meta’s approach to pairing Ember with its more advanced Muse Wonder model allows enterprises to allocate tasks according to their complexity and volume, optimizing overall AI resource use. Given that Ember leads in Meta’s AutomationBench test for workflow completion, it shows promise in real-world business automation scenarios where cost and reliability are critical.
What to watch next
Meta plans to expand the Muse lineup with even more efficient AI models currently in training, potentially reshaping competitive dynamics in AI model pricing and capabilities. Keeping the model weights private for Muse Ember and other frontier models indicates Meta’s strategic focus on proprietary control while balancing openness with prior lower-capacity releases.
Industry observers should monitor how Muse Ember performs in broader external deployments beyond Meta’s internal testing, including safety assessments covering cybersecurity and ethical boundaries. The model’s competitive pricing against notable rivals from both Western and Chinese markets will also be a key factor influencing adoption by startups and larger AI consumers.