Advertise hereReach developers building with AI coding agents. →
Muse Agents

Meta's Muse Ember matches Claude Haiku 5.5's score at about a third of the cost per task: $0.10 per million input tokens

Meta released Muse Ember, its smallest and cheapest Muse model, built for the repetitive work agents do all day: tool calls, coding subtasks, routing, extraction and parsing long documents. It reads text, images, audio, video and PDFs, has a 1M-token context window and can write up to 512K tokens. On the Artificial Analysis Intelligence Index it scores 36, tying Claude Haiku 5.5 at max effort, at about $0.14 per task versus roughly $0.41, Meta says. Meta suggests pairing it with a bigger model: Muse Wonder or Spark plans, Ember handles the volume.

Why it matters

Subagent calls are where agent bills pile up. A model this cheap that holds its own on agent and coding benchmarks can cut the cost of high-volume workflows a lot.

Who it's for

Developers running coding agents, document pipelines or customer workflows who want a cheap, fast subagent model.

How to try it

  1. Create a Meta Model API key, then call the Responses API with model muse-ember-1.0 (Chat Completions and Messages endpoints also work).
  2. Start with a high-volume task, for example classifying support tickets or extracting fields from PDFs, and compare its output and cost with your current small model.
  3. In a multi-agent setup, keep a bigger model as planner and send bounded subtasks to Ember.
  4. Keep requests under 200K input tokens where you can; longer ones are billed at the higher rate in full.

Plan & availability: Available November 9, 2026 on the Meta Model API (muse-ember-1.0, Standard mode only) and in Muse Code on pay-as-you-go and every plan; also on Google Cloud and Oracle Cloud, and in OpenRouter and OpenCode per their announcements. Price: $0.10 input, $0.004 cached input and $0.50 output per million tokens for requests up to 200K input tokens; $0.30/$0.012/$1.50 above that. Benchmark claims are Meta's and Artificial Analysis's; Meta's own chart shows it trailing Haiku 5.5 on FrontierCode 2.0. Checked on research.meta.ai and dev.meta.ai on November 10, 2026.

Share on X

Source: Meta AI on research.meta.ai ↗

Written by AgentStackList from the original post. Checked Nov 10, 2026. Features, plans and availability can change, so confirm on the source before you rely on it.

muse-embermetaapisubagentscheap-models

More Muse tricks

All Muse tricks →

Agent Stack Weekly · free

Get the week's best AI tricks

Limit and pricing changes, new MCP servers, and setup fixes for Claude Code, Codex, Cursor, Copilot and Antigravity.

No spam. Unsubscribe anytime. Read past issues