Key Notes
- Muse Ember is available through Muse Code and Meta Model API for repeated coding, tool use and multimodal document work, with a context window of roughly one million tokens.
- Base API pricing starts at $0.10 per million input tokens and $0.50 per million output tokens, while requests above 200,000 input tokens are billed entirely at the higher long-context rates.
- Meta’s Artificial Analysis comparison shows an intelligence score close to Haiku 5.5 at about one-third of the estimated benchmark task cost, with results varying across individual tests.
Meta has launched Muse Ember, a smaller AI model aimed at the repetitive work inside agent workflows. Starting at $0.10 per million input tokens and $0.50 per million output tokens, it is designed to make tool calls, targeted code changes and document processing economical at scale.
The company introduced Ember on November 9. It is available in Muse Code and through Meta Model API as muse-ember-1.0. The official model page describes support for text, images, audio and video, while Meta’s documentation gives it a roughly one-million-token context window for working across large collections of material.
A smaller worker behind bigger models
Meta proposes dividing a job between models: Wonder or Spark handles planning and difficult decisions, while Ember executes well-defined subtasks. That expands the Muse Wonder lineup into the lower-cost work that agents may repeat hundreds of times, such as reading files, checking outputs and extracting information from documents, including PDFs.
In a developer walkthrough, Meta demonstrates Wonder supervising Ember workers as they migrate 16 syntax-highlighting components from Python to Rust. The company estimates that run at $3.96 under pay-as-you-go pricing. The example illustrates the intended division of labor, with a larger model coordinating many smaller assignments.
The price changes above 200,000 input tokens
Meta’s pricing documentation sets the base rates at $0.10 per million input tokens and $0.50 per million output tokens for requests with up to 200,000 input tokens. Cached input costs $0.004 per million tokens in that tier.
Above that threshold, the rates rise to $0.30 for input, $1.50 for output and $0.012 for cached input. The higher tier applies to the entire request. Cached tokens and the history of a continued, stored conversation count toward the threshold, which makes context management part of the cost calculation.
Meta’s comparison using November 9 Artificial Analysis results puts Ember at 35.8 points against Haiku 5.5’s 36.2, both rounded to 36 in its summary. Estimated benchmark task costs are $0.136 and $0.411 respectively, explaining the claim of roughly threefold savings. Those figures depend on the benchmark, reasoning setting, token usage and list prices.
The results also show differences by task. On FrontierCode 2.0, Haiku scored 34.9% against Ember’s 27.2%. Meta’s evaluation report distinguishes externally sourced results from internal tests and identifies the agent setups used. For teams adopting Ember, the practical decision is which repeatable tasks it can complete reliably enough to justify the lower cost.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
