Meta Muse Ember API Pricing: $0.10 Input, $0.50 Output and the 200K Token Catch

Published:

Meta Superintelligence Labs released Muse Ember on November 9, 2026, its smallest and cheapest Muse model, aimed at the repetitive work AI agents do all day. The list price is $0.10 per million input tokens and $0.50 per million output tokens. That headline number hides a catch: prompts over 200,000 tokens cost three times as much. Here is the full price list, how it compares with Claude Haiku 5.5 and GPT-6 Luna, and where the savings are real.

Meta Muse Ember API pricing

The rates below come from Meta’s pricing documents as reproduced by eesel and confirmed in Fello AI’s launch write-up. The model ID is muse-ember-1.0.

  • Up to 200K input tokens: $0.10 per million input, $0.004 per million cached input, $0.50 per million output.
  • Over 200K input tokens: $0.30 input, $0.012 cached input, $1.50 output.
  • Reasoning tokens are billed at the same rate as visible output.
  • Web search: $2.50 per 1,000 queries, per eesel.
  • Batch: no batch tier is listed.

The 200K threshold counts input tokens including cached ones, and a request that continues a stored conversation counts that history too. Once a request crosses the line, the higher rate applies to the whole request, not only the part above 200K. Fello notes that Meta’s launch post quotes only the lower rate, while the developer page lists both.

How Muse Ember compares with Claude Haiku 5.5 and GPT-6 Luna

On the standard price list, Ember is not cheaper. Both sources show Claude Haiku 5.5 and GPT-6 Luna at the same $0.10 and $0.50. The differences are in the details:

  • Cached input: $0.004 for Ember against $0.01 for Haiku 5.5 and Luna.
  • Long prompts: Haiku 5.5 rises to $0.50 input and $2.50 output above 100,000 tokens, per Fello, so Ember’s step at 200K is later and smaller.
  • Batch: eesel lists a 50% batch discount for Haiku 5.5 and a batch option for Luna. Ember has none, so overnight bulk jobs may cost less elsewhere.
  • Default effort: Ember defaults to the maximum reasoning effort when you do not set one, and the lowest setting is minimal. Setting a lower effort for simple tasks such as classification is the main way to control cost.

Cost per task is where Meta makes its case. Artificial Analysis, as cited by Fello, puts the cost to run its Intelligence Index at about $0.14 per task for Ember, $0.41 for Haiku 5.5 and $0.15 for Luna, against $2.39 for Meta’s larger Muse Spark 1.3. Fello also reports that Ember produced 115 million output tokens on that run against 298 million for Haiku 5.5, and that Artificial Analysis calls it somewhat verbose for its price class.

What the benchmarks show

Fello lists results for Ember, Haiku 5.5 and Luna in that order. Ember leads on several: AutomationBench-AA at 37.8 against 13.9 and 20.2, the Coding Agent Index at 46 against 37 and 31, and Terminal-Bench 4.0 at 40.9 against 39.2 and 16.4. Haiku 5.5 wins FrontierCode 2.0 at 34.9 against Ember’s 27.2. On the Artificial Analysis Intelligence Index, Ember and Haiku 5.5 tie at 36, and Luna scores 28.

Treat the table with care. Fello says Meta ran four of the ten rows itself and calls the competitor runs “best-effort”, and some figures are Meta’s own composites. Independent testing on your own workload will tell you more than a vendor chart.

Who can use it and how

  • It is available on the Meta Model API and in Muse Code. Fello says the API works with the OpenAI SDK and supports the Responses API.
  • It is not in the Meta AI app, WhatsApp or the Muse agent, according to Fello. The weights are not released.
  • eesel lists default rate limits of 3,000 requests and 4 million tokens per minute per team.
  • Muse Code plans run from $5 to $200 a month. eesel says Ember is included in every plan; Fello says the November 9 plan cards did not list it, so confirm in Meta’s documentation.
  • Muse Code plans default to a mode that lets Meta train on prompts and code unless the user turns it off. eesel says that mode is not available for Ember on the API.

Fello reports a 1 million token context window and up to 512,000 output tokens. eesel says the page it read did not state either figure, so verify before you design around them.

When the price works, and when it does not

The cost savings are real for high-volume, bounded jobs: tagging, routing, extraction and short agent steps that stay under 200K tokens. eesel’s own estimate for triaging 20,000 support tickets a month is about $5 at minimal effort and about $22 at maximum effort, using its token assumptions, so treat that as an illustration rather than a quote. The model is a weaker choice for long uncapped agent sessions that drift past 200K tokens, for batch-heavy work, and for hard coding tasks where Haiku 5.5 scores higher on FrontierCode.

For other recent model price changes, see our posts on the Gemini Decisions API and on the Claude Fable 5.2 API price cut. Prices change often, so check Meta’s pricing page before you budget.

Sources

We did not open Meta’s own launch post. eesel sells a helpdesk agent, so its support-team examples are illustrative. Image: Meta Platforms headquarters in Menlo Park, photo by LPS.1, CC0, via Wikimedia Commons.

Like our coverage? Add stepphase.com as a preferred source on Google to see more of our stories in Top Stories.

Saif Ali Tai
Saif Ali Taihttps://xplorely.com/
I'm Saif Ali Tai, a tech blogger and software engineer with 8+ years of experience building apps, and I run StepPhase as its editor. I cover technology news, AI tools and security stories that matter to everyday users. Articles here are drafted with AI assistance and checked against the sources linked in each post.

Related articles

Recent articles