Meta's Muse Ember costs 14 cents per task, undercuts Claude Haiku

Meta's new AI model, Muse Ember, completed benchmark tasks for about 14 cents each — roughly a third of what Anthropic's Claude Haiku 5.5 cost at 41 cents, despite identical standard token rates. The Superintelligence Labs team introduced the model on Nov. 9, calling it Meta's smallest and most cost-efficient AI model yet.

The price gap matters because of what Ember is for. It's built for the steps an agent might repeat dozens of times in a single assignment: pulling details from a document, calling another tool, or running a test after a code change. Big models can plan the job; someone still has to pay for every small step underneath.

What does this change for developers?

Ember slots into Meta's Muse model family as the budget tier. Meta positions Muse Wonder for the hardest problems and Muse Spark for coding and tool-heavy work. The pitch: developers use Wonder to plan a complicated job and hand the individual steps to Ember. For now, Ember is available through Meta's Muse Code tool and its developer API — a limited distribution, and Meta hasn't said when that widens.

The model can take in up to a million tokens at once, enough to work through large document sets. It also handles images, audio, and video alongside text, so an agent could read a scanned contract or pull details from a meeting recording. In early testing Meta shared, cloud storage company Box used Ember to check contracts clause by clause against its negotiation rules and assign an overall risk rating.

How does the pricing actually stack up?

On Artificial Analysis's benchmark, Ember and Claude Haiku 5.5 both rounded to a score of 36 at their highest reasoning settings. Both charge 10 cents per million input tokens and 50 cents per million output tokens — yet Ember's per-task cost came in at about 14 cents versus Haiku's 41. OpenAI's GPT-6 Luna scored 28 and cost about 15 cents per task at its highest setting.

Meta's own published results are a mixed bag, worth reading closely. Ember beat Haiku on a terminal coding test, but Haiku did better on a test judging whether code changes were ready to merge. Meta also said it already has even cheaper models in training, which cuts both ways: good news for costs, less so for anyone anchoring a stack to Ember today.

Why the cheap tier matters

Ember exists because of the Muse app, Meta's AI assistant with the Jolly avatar, which the company has promoted since September. The app passed 5 million downloads in 22 days, and Meta started rolling out voice conversations and animated avatars in the U.S. last week. Muse can work across some apps with a user's permission.

The bet behind all of it: AI assistants will take on everyday chores like booking flights, where sorting dozens of fares — each with its own carry-on rules — means dozens of cheap model calls. If Meta wants people handing off those tasks daily, the little steps have to be cheap on its end too. Expect that cost pressure, and the cheaper models behind it, to keep arriving.