AI modelsNovember 10, 2026
27
š¶ļø
Meta Ember: Cheap Agent Crushes Haiku
Meta's proprietary Muse Ember model reportedly matches Claude Haiku 5.5 at roughly one-third the estimated per-task cost.
#Meta#Muse Ember#LLM#pricing#inference
š„ What happened
Meta dropped Muse Ember on Monday ā a tiny, cheap model built for agentic grunt work. It scores 36 on Artificial Analysis's Intelligence Index, matching Claude Haiku 5.5, but costs 14 cents per benchmark task versus Haiku's 41 cents.
š” Why it matters
Ember burns only a third of Haiku's output tokens at max reasoning and completes 37.8% of AutomationBench tasks on the first try ā nearly double GPT-6 Luna's 20.2%. The catch: above 200K input tokens, Meta triples the price, making Ember pricier than some rivals for long-context jobs.
ā” Our take
Meta just copied DeepSeek's pricing playbook ā and that's a wake-up call for Anthropic and OpenAI. If you run agents at scale with short prompts, Ember is hard to ignore.
The title, summary and analysis of this item were produced automatically by an AI system and have not been editorially reviewed. They may contain errors, bias or omissions ā when in doubt, read the linked original source.
