Our article2 min read
Meta Muse Ember: A Cheap AI Model, the 200K Price Catch and Who Can Use It
Meta launched Muse Ember on 9 November, a small low-cost model for agent work. The headline price rises for long prompts, and the benchmarks are largely Meta's own.
In this article
Key takeaways
- Meta calls Muse Ember its smallest and most efficient model, offered through the paid Meta Model API and Muse Code.
- The headline price is $0.10 per million input tokens and $0.50 per million output tokens, but it rises above 200,000-token prompts.
- Meta ran some of the benchmark comparisons itself and calls its competitor runs best-effort.
- It is a hosted API product, not an app for everyday chat.
What Meta released
On 9 November 2026 Meta introduced Muse Ember, which it describes as its smallest and most efficient model. Meta says it is built for agent-style work such as calling tools, small coding jobs, routing requests, pulling data out of text and reading long documents. It has a one-million-token context window, which is roughly the amount of text a model can consider at once, and Meta says it can produce up to 512,000 tokens of output.
Price, and the catch
Meta lists $0.10 per million input tokens and $0.50 per million output tokens, with cheaper cached input at $0.004. Those match the list prices felloai gives for Claude Haiku 5.5 and GPT-6 Luna on short prompts.
The catch, reported by felloai, is that prompts longer than 200,000 tokens cost more: $0.30 per million input tokens and $1.50 per million output tokens, according to Meta's own pricing table. Meta's launch post quotes only the lower rate. Felloai says that even the higher tier is still below Haiku 5.5's long-prompt price.
What the numbers show
Meta reports a score of 36 on the Artificial Analysis Intelligence Index, which it says ties Claude Haiku 5.5 and beats GPT-6 Luna at 28. On one agent benchmark, AutomationBench-AA, Meta reports 37.8 percent against 13.9 percent for Haiku and 20.2 percent for Luna. Meta also estimates about $0.14 per task on the Artificial Analysis index.
Not every row favours Ember. Meta's own table shows Haiku 5.5 ahead on FrontierCode 2.0, 34.9 percent to 27.2 percent. Meta also says it matches Haiku 5.5's best score on that index at about a third of the cost, and it quotes early testers such as Cloudflare, which reported up to 30 percent faster tail latency. As noted below, those comments were picked by Meta, so they are best read as a company's own account.
Reasons for caution
- Meta ran four of its ten benchmark rows itself and, according to felloai, called the competitor runs best-effort and not necessarily tuned for proprietary models.
- Felloai says one benchmark row mixes three different test setups, so it is not a clean comparison.
- Meta says Ember handles text, images, audio and video, but felloai says Artificial Analysis lists only text and image inputs.
- The partner comments in Meta's post were chosen by Meta.
Who can use it
Meta's page presents Muse Ember as a hosted API product through Muse Code and the Meta Model API, under the name muse-ember-1.0, and does not say whether the weights are open. Felloai states it is proprietary and not in the Meta AI app. Meta also says Muse Ember is available on Google Cloud and Oracle Cloud.
What it means for you
Most readers will never call this model directly. The news matters because small, cheap models do the repetitive work behind many apps, and lower prices can reach the products you use. If you build with AI, test Ember on your own tasks and check your typical prompt length before trusting the headline price. We have not tested the model ourselves.
Go deeper
Sources
Written by an AI from the sources above and checked by a person before publishing. We have not tested products unless we say so. See our editorial policy or tell us about an error through contact.
