Meta launches Muse Ember, a low-cost model for AI agent workflows

Meta announced Muse Ember on November 9 for the repetitive tasks performed by AI agents. The company describes it as its smallest and most efficient model, now available through the Meta Model API.

Dji Banner Inside Article - 728x90

Built for repeated tasks

Ember focuses on tool calls, coding subtasks, routing, extraction and long-document processing. Meta proposes pairing it with larger Muse models, assigning complex planning to a larger model and frequent execution steps to the smaller one.

Meta’s product page lists text, image, audio and video inputs. A one-million-token context window accommodates long documents and extensive task context, although capacity alone does not guarantee accurate understanding.

API pricing changes with context length

Meta lists two tiers, with all prices in US dollars per million tokens:

• Below 200,000 tokens of context: $0.10 input, $0.50 output and $0.004 cached input.
• Above 200,000 tokens of context: $0.30 input, $1.50 output and $0.012 cached input.

The headline starting rate therefore does not apply to every long-context request. These are API usage rates, separate from Muse Code subscriptions.

Vercel confirms availability

Vercel also announced Ember’s arrival on AI Gateway on November 9. The model identifier there is meta/muse-ember. Vercel lists text, image and video inputs, adjustable reasoning from minimal to max, and support for Chat Completions, Responses and Anthropic Messages APIs.

Reading the performance claims

Meta’s comparison combines agent, coding and multimodal tests. It does not show Ember leading on every task: the FrontierCode 2.0-AA row gives Ember 27.2 and Claude Haiku 5.5 a score of 34.9. Developers should measure accuracy, latency and total cost in their own workflows rather than treating a vendor’s benchmark table as a universal ranking.

Sources: Meta announcement and evaluations, Meta specifications and pricing, Vercel availability announcement.

Image: Meta’s official Muse Ember product page.