Muse Ember: Meta's $0.10 per 1M input token agent model

Written by , Founder of Launch Llama · November 10, 2026

TL;DR
  • Meta released Muse Ember on Monday, its lightest and most cost-efficient model so far, aimed at tool calls, coding subtasks, routing, extraction, and long document parsing.
  • It costs 10 cents per million input tokens, which Meta says is cheaper than Anthropic's Haiku and OpenAI's Luna and cost-competitive with DeepSeek V4.1 Flash and GLM 5.3 Flash.
  • Ember handles text, images, and audio, and is designed to pair with the larger Muse Wonder: Wonder plans, Ember handles the volume.

What happened

Meta Superintelligence Lab released Muse Ember on Monday. Meta calls it the lab's lightest and most cost-efficient model yet. It says Ember is "designed for complex reasoning execution and the work agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing."

Meta leads with cost per task. Using Artificial Analysis's composite benchmark score, it shows Ember as cheaper and smarter than Anthropic's Haiku and OpenAI's Luna. Input pricing is 10 cents per million tokens. That is also cost-competitive with open source options like DeepSeek V4.1 Flash and GLM 5.3 Flash. These comparisons come from Meta's own release, and the source does not give the underlying scores.

Ember is meant to run alongside a larger model such as Muse Wonder in an agentic harness. In a customer-support example, the heavier model sets the rules for an issue, then Ember applies them across tens of thousands of cases. Ember also takes text, images, and audio, so agents can read screenshots, charts, and scanned documents, or work from voice memos, meeting recordings, and video without a separate model. It is the fourth Muse model, after Spark (April), Glimmer (August), and Wonder (October).

What this means for founders

  • Cheaper high-volume agent work. At 10 cents per million input tokens, repetitive steps like extraction, routing, and parsing could cost less to run. The source gives only the input price. Output pricing is not confirmed yet, so check it before you do the math.
  • Planner plus worker is the pattern. Meta's own framing is a larger model for planning and a light model for volume. If your agent uses one expensive model for everything, this is a prompt to split the work.
  • One model for mixed inputs. If your product handles screenshots, scanned documents, or voice memos, Ember could replace a separate model for each type. That would simplify your stack, but test it on your own data first.
  • Benchmarks are Meta's. The cost and quality comparison uses a composite score Meta chose to show. Run your own tasks before you switch anything.
  • Availability is unclear. The source does not say where or how developers can access Ember, so confirm that before you plan a migration.

What the Launch Llama directory shows

Live numbers from Launch Llama for listings that describe themselves as "ai agent", "llm api", "small language model", "agent execution" tools, as of 2026-11-10.

SignalLaunch Llama data
Listings in this space322
New in the last 30 days45
With a free or freemium plan82%
Most upvoted by foundersKiloClaw (774 upvotes)

Launch Llama lists 322 matching listings in this space, 45 of them live in the last 30 days, and 82% have a free or freemium plan, so cost-focused agent tooling is crowded and still growing fast.

How to try it today

  1. Read Meta's Muse Ember announcement and check how developers can access the model. The source does not confirm this yet.
  2. Pick one repetitive step in your product, such as extraction, routing, or document parsing, where cost per task matters.
  3. Collect 50 to 100 real examples of that task, including any screenshots or audio you handle.
  4. Run the same examples through your current small model and through Ember. Compare accuracy and total cost, including output tokens once pricing is confirmed.
  5. If Ember holds up, keep your larger model for planning and send only the high-volume steps to Ember.
Task: Extract {fields} from the attached document.
Return JSON only. If a field is missing, return null.
Document: {paste sample here}

Related tools on Launch Llama

Tools from the Launch Llama directory that fit with this news:

ToolWhat it doesWhy it matters here
Router by RampFor engineers: Router by Ramp routes every LLM request to the lowest-cost model that meets your needs so you cut inference costs and simplify API management.Router by Ramp sends each LLM request to the lowest-cost model that meets your needs, which fits the same cost-per-task thinking behind Ember.
BaarFor LLM agent builders: Baar enforces runtime cost limits before API calls execute, preventing runaway token spend and unexpected bills from agent loops.Baar enforces runtime cost limits before API calls run, a useful guard when agents repeat tasks at high volume.
One API for the World’s Leading AI ModelsFor developers: API-Route provides one OpenAI-compatible gateway to GPT, Claude, Gemini, and other models so you switch AI providers without code changes.API-Route is an OpenAI-compatible gateway for switching between model providers, which helps if you want to compare cheaper models without code changes.

Browse all tools or list your own tool.

Llama's take

I would test Ember this week if you run an agent with a lot of repetitive steps, like support triage, extraction, or document parsing. A 10 cent input price is worth a side-by-side run, especially if you now send everything to a bigger model.

I would wait if you need firm numbers. The source gives no output pricing, no access details, and no raw benchmark scores. The comparison against Haiku and Luna is Meta's own claim.

The thing to watch is how it performs next to Muse Wonder in real agent workflows, because that planner and worker split is the whole pitch.

FAQ

What is Muse Ember?

Muse Ember is a new model from Meta Superintelligence Lab, described as its lightest and most cost-efficient yet. It is built for repetitive agent work like tool calls, coding subtasks, routing, extraction, and long document parsing.

How much does Muse Ember cost?

Meta lists Ember at 10 cents per million input tokens. The source does not give output token pricing, so the full cost per task is not confirmed yet.

Can Muse Ember handle images and audio?

Yes. Meta trained Ember to work across text, images, and audio. It says agents can read screenshots, charts, and scanned documents, or work from voice memos, meeting recordings, and video without a separate model.

Sources

Related guides

Explore founder tools

Free calculators, plus newsletters and launch directories curated for founders.

Mini Apps →Newsletters →Launch Directories →

For bots & AI agents

Machine-readable versions of this article and Launch Llama. Fetch these instead of parsing this page.

This article (.md) llms.txt llms-full.txt ai.txt Full catalog (.md) Endpoint index API docs MCP server