← All Articles
Sign in
Pricing Blog

Muse Ember API Pricing: $0.10 per Million Input Tokens and the 200K Step That Triples It

By AI SaaS Radar Team · Nov 2026 · 4 min read

Meta released Muse Ember on November 9, 2026, priced at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 200K tokens, and three times that above 200K. It is Meta's smallest Muse model, aimed at the cheap, high-volume steps inside agents (tool calls, coding sub-tasks, extraction, document parsing), and it matters most to teams paying for those steps with a mid-tier model today.

Key facts

What the price means for a real workload

Take a sub-agent that reads 10 million input tokens and writes 2 million output tokens a day, every prompt under 200K tokens. On Muse Ember that is $1.00 of input plus $1.00 of output, so $2.00 a day. The same volume on Muse Spark 1.3, which Meta lists at $1.25 input and $4.25 output, is $12.50 plus $8.50, so $21.00. That is our arithmetic on Meta's published rates, before any caching.

Caching pushes the gap further. Cached input at $0.004 per million is one twenty-fifth of the base input rate, so an agent that resends the same system prompt and tool definitions on every call pays very little for them once they are cached.

The 200K step is the line to watch

The 1M context window is the headline, but the price triples once a prompt passes 200K tokens. The same 10M in, 2M out day costs $6.00 if every prompt runs above that line. Long-document parsing is one of the jobs Meta pitches Ember for, so a pipeline that stuffs whole contracts or codebases into one prompt can land on the higher rate without anyone deciding it should. Splitting documents into chunks under 200K keeps the lower rate, at the cost of more calls.

Read the benchmarks as Meta's numbers

Every comparison in the launch post is published by Meta, and several of the benchmarks are Meta internal. On Meta's table at max reasoning, Ember scores 40.9% on Terminal-Bench 4.0 against 39.2% for Claude Haiku 5.5 and 16.4% for GPT-6 Luna, and 37.8% on AutomationBench-AA against 13.9% and 20.2%. Haiku 5.5 leads on FrontierCode 2.0-AA, 34.9% to Ember's 27.2%. On cost per Intelligence Index task, Meta's chart puts Ember at $0.136 for a score of 35.8 and Haiku 5.5 at $0.411 for 36.2, which is where the "about a third of the cost" claim comes from.

Two cautions. The cost-per-task figures are at max reasoning, so your bill depends on the effort setting you run. And the launch page's own table and chart round the Haiku comparison differently (36 against 36.2), which is a reminder to test on your own tasks before moving traffic. We covered why vendor benchmarks rarely settle a model choice in our piece on SWE-bench and MMLU.

Who should test it this week

Ember fits the planner and executor split Meta describes: a larger model plans and a small one handles the volume. If your agents already route sub-tasks to a cheap model, Ember is a candidate for that slot. If they run one model for everything, the saving only shows up after you split the work, and that split has its own costs, as the multi-agent tax shows. For Meta's wider small-model strategy, see our Muse Spark analysis.

What to do

More pricing changes are tracked in the article archive.

Questions people ask

How much does Muse Ember cost?

$0.10 per million input tokens, $0.004 per million cached input tokens and $0.50 per million output tokens for prompts up to 200K tokens. Above 200K the rates are $0.30, $0.012 and $1.50.

Is Muse Ember better than Claude Haiku 5.5?

Meta's own results put Ember ahead on most agent benchmarks it reports and behind Haiku 5.5 on FrontierCode 2.0-AA, with similar Intelligence Index scores at about a third of the cost per task. These are Meta's figures, not independent tests.

Where can I use Muse Ember?

On the Meta Model API as muse-ember-1.0, in Muse Code, on Google Cloud Model Garden, and through OpenRouter, which lists Google Vertex and Meta as providers.

Does Muse Ember have open weights?

Meta has not published weights or a parameter count for Ember. It is available through the API and partner clouds.

Sources

Stay ahead of the AI SaaS market

Sourced, dated analysis on security, funding, and benchmarks. Straight to your inbox.

No spam. Unsubscribe anytime.