Meta Unveils Lightweight Muse Ember Model Targeting Cost-Efficient Agent Workflows
Frontier AI models · AI assistants
Meta has announced its lightweight Muse Ember model, featuring a 1 million token context window and up to 512K output tokens with multimodal support. Meta claims it matches Claude Haiku 5.5 benchmark performance at one-third the task cost via its API.
Meta announced the lightweight Muse Ember model on November 9, designed for complex reasoning and repetitive agent tasks. The model features a 1 million token context window and up to 512K output tokens, offering native multimodal understanding across text, images, and video to handle tool calling, routing, coding, and long-document processing. According to Meta, Muse Ember is the first model trained using Muse Wonder Full. Available via the Meta Model API, pricing is set at $0.1 per million input tokens and $0.5 per million output tokens, with cached inputs priced at $0.004 per million tokens. Meta's benchmark figures indicate Muse Ember matches or leads Claude Haiku 5.5 across several agent and coding tasks, while trailing in physical engineering and complex code merging. Meta highlights its cost-performance ratio, stating partners like Box are testing it for contract review and code generation.
Key takeaways
- Muse Ember features a 1 million token context window and 512K output tokens, tailored for agent workflows, tool calling, and long documents.
- Meta claims the model delivers comparable benchmark performance at one-third the task cost of Claude Haiku 5.5 and surpasses GPT-6 Luna settings.
- API pricing is set at $0.1 per million input tokens and $0.5 per million output tokens, with cached input at $0.004.
Source：iThome Read the original ↗
Prepared, checked and automatically published by AI; not individually reviewed by a human. Translation checked by AI.