Meta Launches Lightweight Model Muse Ember, Benchmark Scores Rival Claude Haiku 5.5, Cost Per Task Only 1/3
Meta has launched Muse Ember, its smallest and most efficient model to date, designed specifically for complex reasoning and everyday agent tasks. Meta says it performs comparably to Claude Haiku 5.5 on the Artificial Analysis leaderboard, at only 1/3 the cost per task, and surpasses GPT-6 Luna on multiple tests. The model has a 1 million Token context window, outputs up to 512,000 Tokens, natively supports text, images, audio and video, and is available starting today on Meta Model API, Google Cloud and Oracle Cloud, at $0.10 per million input Tokens and $0.50 per million output Tokens.
Why read this
AI score 74
Meta's smallest model to date offers benchmark and cost comparisons with Claude Haiku 5.5, which can be used to gauge where lightweight models stand in terms of cost-effectiveness.
Topics
Body · AI translation
Meta Launches Lightweight AI Model Muse Ember: Benchmark Scores Rival Claude Haiku 5.5, Cost Per Task Only 1/3
2026-11-10 08:59 IT Home - Luyuan
English text is being prepared.
IT Home, November 10 news: Meta has now launched the Muse Ember AI model. This is Meta's smallest and most efficient model to date, designed for executing complex reasoning tasks and for the repetitive day-to-day work handled by agents.
Meta says that on the Artificial Analysis leaderboard, this model's performance rivals Claude Haiku 5.5, but its cost per task is only 1/3, and it surpasses GPT-6 Luna in multiple tests.
AutomationBench-AA (end-to-end business workflows)
Agents' Last Exam (end-to-end agent tasks)
Personal Superintelligence Index (internal composite index for personal and knowledge work)
Terminal-Bench 4.0 (agent terminal programming)
Coding Agent Index-AA (programming in real agent environments)
FrontierCode 2.0-AA (real engineering tasks evaluated by code merge feasibility)
At the same time, this model has a 1 million Token context window and supports up to 512,000 output Tokens, and can process an entire codebase or a full set of documents in a single call. It allows agents to execute hundreds of programming subtasks in a single session, including reading files, running tests, and making targeted fixes.
Muse Ember natively supports understanding text, images, audio, and video. Agents can read screenshots, charts, and scanned documents without relying on other models, and can also process voice memos, meeting recordings, and videos. It can handle long real-world documents, operate a computer directly based on screen screenshots, and recognize subtle visual details.
On the security side, this model was evaluated under Meta SI's extended framework before release, and all test results show that Muse Ember's risk level is below the high-risk threshold set by the framework.
On pricing, this model is available starting today on Meta Model API, Google Cloud, and Oracle Cloud. The input price is $0.10 per million Tokens (IT Home note: about 0.67 yuan at current exchange rates), and the output price is $0.50 per million Tokens (about 3.3 yuan at current exchange rates).