Muse Ember: Meta's smallest AI model charges $0.10 per million tokens
By Anika SharmaAI & Innovations Editor, Peepals Tech
Representative image: pixel2013 / Pixabay
Meta released Muse Ember on 9 November 2026, a small AI model for repetitive agent tasks priced at $0.10 per million input tokens, with a one-million-token context window.
What Happened
Meta released Muse Ember on Monday, 9 November 2026, describing it as the smallest and most efficient model in its Muse family of AI models. Built by Meta Superintelligence Labs, the company's AI research unit, the model is meant for the high-volume, repetitive work that AI agents do, such as calling software tools, handling small coding tasks, routing requests and pulling data out of long documents. Meta prices it at $0.10 per million input tokens and $0.50 per million output tokens, and it is available only through Meta's paid services, not as a download.
Key Facts
- Launch date: 9 November 2026, announced on Meta's AI research blog; the model ID for developers is muse-ember-1.0.
- Price: $0.10 per million input tokens and $0.50 per million output tokens. A token is a small chunk of text, roughly part of a word, that AI companies use to measure usage.
- Context window: one million tokens, meaning the model can take in a very large amount of text at once; reports say it can write up to 512,000 tokens in one reply.
- Inputs: Meta says it natively handles text, images, audio and video.
- Access: through the Meta Model API and Muse Code; the weights (the trained model files) stay private, unlike Meta's Muse Glimmer and Muse Spark Open.
- Meta's benchmark claim: 37.8% on AutomationBench, a test of automated multi-step tasks, against 13.9% for Anthropic's Claude Haiku 5.5, according to the company.
Why It Matters
Most money spent on AI by businesses is not spent on one clever answer but on millions of small, routine calls. An AI agent, a program that carries out tasks step by step on its own, may make dozens of model calls to finish one job: checking a database, filling a form, reading a file. Each call costs money. A cheap, fast model for that routine work lowers the bill for running agents at scale, and that is the market Meta is aiming at with Ember.
Meta describes a division of labour inside its own range. Its larger Muse Wonder model is meant to plan, while Ember is meant to carry out the volume of small steps. Meta says Ember is the first model trained with Muse Wonder Full, its bigger system. This mirrors how other AI companies now sell pairs of models: a large one for reasoning and a small one for repetitive execution.
The benchmark figures should be read with care. The AutomationBench score comes from Meta's own announcement. One independent review site reports that Ember and Claude Haiku 5.5 both score 36 on the Artificial Analysis Intelligence Index, a combined ranking run by an outside group, with Ember costing less per average task. That suggests similar overall ability, with Meta's advantage claimed mainly on price and on agent-style tasks.
For Indian developers and start-ups building agents for customer service, document handling or coding help, prices per million tokens decide whether a product is affordable to run. Ember is priced in US dollars and is reached through Meta's cloud service; because the weights are not released, it cannot be run on a company's own servers, which matters for firms that must keep data in-house.
Peepals calculation: Under Meta's prices, an agent job that reads 10 million tokens and writes 2 million would cost $1.00 for input (10 × $0.10) and $1.00 for output (2 × $0.50), or $2.00 in total. On Meta's own AutomationBench figures, Ember scored about 2.7 times Claude Haiku 5.5 (37.8 ÷ 13.9 ≈ 2.72). Our calculation from the reported figures.
| Item | Muse Ember (as reported) |
|---|---|
| Released | 9 November 2026 |
| Input price | $0.10 per million tokens |
| Output price | $0.50 per million tokens |
| Context window | 1 million tokens |
| Weights | Not downloadable |
Impact
Short-term: Developers already using Meta's API can switch routine agent steps to Ember to test whether it cuts their costs without hurting results.
Long-term: Price competition among small models is pushing the cost of each AI step down, which makes agents that run many steps cheaper to operate and more common in business software.
Who is affected: Software developers and start-ups building AI agents, companies comparing AI suppliers on cost, and rival AI firms selling small models.
Key Takeaway
Meta's Muse Ember, launched on 9 November 2026, is a small, closed AI model priced at $0.10 per million input tokens for routine agent work.
Questions and Answers
What is Muse Ember?
Muse Ember is the smallest and cheapest model in Meta's Muse family, released on 9 November 2026 for repetitive AI agent tasks such as tool calls, data extraction and routing.
How much does Muse Ember cost?
Meta charges $0.10 per million input tokens and $0.50 per million output tokens through its Meta Model API.
Can I download Muse Ember and run it on my own computer?
No. Its weights remain private, so it is available only through Meta's API and Muse Code, unlike Meta's Muse Glimmer and Muse Spark Open.
Is Muse Ember better than Claude Haiku 5.5?
Meta says Ember scored 37.8% against 13.9% for Claude Haiku 5.5 on AutomationBench, but that is the company's own test; an independent index reported by a review site puts the two level at 36.
Disclaimer: Prepared by the Peepals newsroom from publicly available sources with AI assistance. Information is accurate to the best of our knowledge at the time of publication and may change. Images may be representative. Not professional advice. Report an error via our contact page.
By Anika Sharma, AI & Innovations EditorCovers AI & Innovation and more for Peepals Tech. All stories by Anika Sharma
Get the day's stories in your inbox
Peepals Weekly: the week decoded, free. One email a week, unsubscribe in one click.
Anika Sharma is a pen name of the Peepals Global Editorial Team, which researches, fact-checks and publishes every story. The portrait is an illustration.
