Meta official logo (public domain, Wikimedia Commons) — CryptoBriefing brand treatment
The newest member of the Muse family targets cost-conscious developers on the Meta Model API
Meta has a new AI model, and its main selling point is how little it is. The company has launched Muse Ember, its smallest reasoning model to date, on the Meta Model API.
The price tag is the headline act: $0.10 per million input tokens and $0.50 per million output tokens. In a market where AI bills can balloon quickly, that is a deliberate pitch to developers who watch every line on their cloud invoice.
What Meta actually shipped #
Muse Ember slots in at the bottom of the Muse lineup, a family of multimodal reasoning models built by Meta Superintelligence Labs.
Ember charges $0.10 for every million tokens fed into it. It charges $0.50 for every million tokens it produces in response.
The broader Muse family is designed around tool use and handling mixed inputs. That includes text, images and audio, which means the models are meant to do more than chat.
Ember sits beneath larger siblings like Muse Spark and Muse Wonder. Those bigger models are aimed at heavier workloads, the kind of jobs where raw capability matters more than cost per query.
How the pricing stacks up inside the Muse family #
The clearest benchmark for Ember is Meta’s own catalog. Muse Spark 1.3 is priced at approximately $1.25 per million input tokens and $4.25 per million output tokens.
Run the math and the gap is wide. On input, Spark 1.3 costs roughly 12.5 times what Ember charges. On output, it costs about 8.5 times as much.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
There is a wrinkle in how Ember’s pricing lines up with the rest of Meta’s API. Its rates align closely with the discounted “Contributor” tier of the Meta Model API.
That tier offers developers meaningful savings. The catch is that they agree to let Meta use their interaction data to train future models.
What has not surfaced yet is a performance report card. No independent benchmarks or technical specifications for Ember have appeared publicly, so it remains unclear how it measures up against Spark, Wonder or rival small models.
Background: Meta’s pivot to a tiered AI shop #
The Muse family is a fairly recent creation. It began with Muse Spark, which launched in April 2026.
Ember arrives as Meta’s smallest reasoning model as of October 2026. That gives the lineup a low-cost entry point roughly six months after the first Muse release.
Since around 2025, Meta has been restructuring its AI efforts, a push that included establishing Superintelligence Labs. That restructuring has tilted the company toward proprietary models and tiered access.
What this means for developers and the AI market #
At $0.10 per million input tokens, reasoning capability becomes a lot more accessible to teams that may have been priced out of higher-end models.
The Contributor-style pricing also feeds Meta’s training pipeline. If a large share of Ember users opt into the discounted, data-sharing arrangement, Meta gains a stream of real-world interaction data to improve future models.
Without published benchmarks, developers have little hard evidence about where Ember excels or falls short. Companies in regulated industries, or those with strict customer privacy commitments, may find the data-sharing condition tied to Contributor-tier savings hard to accept.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.