Meta's smallest and most efficient multimodal model, built for high-frequency agent tasks like tool calling, routing and data ex
Meta released Muse Ember, its smallest and most efficient model, positioned against Claude Haiku and GPT-6 Luna for agent workloads such as tool calls, coding subtasks, routing, extraction and long-document parsing. Short-prompt pricing matches rivals at $0.10 per million input and $0.50 output tokens, but long-context tiers rise more steeply, reaching $0.30 input and $1.50 output beyond 200k tokens.
Meta released Muse Ember, its smallest and most efficient model to date, targeting high-frequency agent workloads such as tool calling, coding subtasks and long-document parsing. On the Artificial Analysis intelligence index it scores about 36 at top reasoning effort, matching Claude Haiku 5.5 and beating GPT-6 Luna, while costing roughly one-third per task at about $0.136, placing it on a favorable efficiency frontier. It is available now via Meta Model API and Muse Code, with more efficient models reportedly in training.
Meta released Muse Ember, its smallest model, matching Claude Haiku 5.5 at 36 on the Artificial Analysis Intelligence Index at roughly $0.136 per task versus $0.411 for Haiku. Positioned as the execution layer in a two-model agent split alongside the planning-focused Muse Wonder, it natively handles text, images, audio and video with a 1M-token context and up to 512K output tokens. The trade-off is weaker real-repo mergeability on FrontierCode 2.0-AA, and several results are Meta internal runs.
Meta has launched Muse Ember, its smallest and most efficient model to date, purpose-built for complex reasoning and high-frequency agentic workloads. It matches Claude Haiku 5.5 on the Artificial Analysis Index (36) at one-third the per-task cost, while leading on agentic and multimodal benchmarks such as AutomationBench-AA and OSWorld 2.1. The model is available immediately via Meta Model API, Google Cloud and Oracle Cloud, priced at $0.10/$0.50 per million input/output tokens.
Meta released Muse Ember, its smallest model to date, claiming intelligence on par with Claude Haiku 5.5 at roughly one-third the cost. It targets token-heavy agent workloads such as tool calls, coding subtasks, routing, extraction and long-document parsing, and is available now on the Meta Model API and in Muse Code. Pricing is $0.10 input and $0.50 output per 1M tokens up to 200K tokens, rising to $0.30 and $1.50 beyond that.
Meta has launched Muse Ember, its smallest and most efficient model yet, purpose-built for the routine work AI agents do all day: tool calling, small coding tasks, request routing, data extraction and long-document processing. Meta claims it delivers more intelligence per dollar at every reasoning level than Claude Haiku 5.5 or GPT-6 Luna, at roughly $0.14 per task, though Haiku still leads on real coding benchmarks. The multimodal model handles up to one million tokens and is available now via the Meta Model API and Muse Code at $0.10 per million input tokens and $0.50 per million output tokens.
Meta has released Muse Ember, an efficiency-tier model that scores 36 on the Artificial Analysis Intelligence Index, matching Claude Haiku 5.5 (max) while costing about one-third as much at $0.14 per task. It scores 68.1% on AutomationBench-AA, level with Meta's flagship Muse Wonder, though it trails the flagship by 8 index points at roughly 1/50th the cost per task.
Meta's Muse Ember is now available on OpenRouter, positioned as a sub-task execution model for coding agents. The suggested pattern uses a frontier planner such as Muse Wonder to decompose work, then fans bounded sub-tasks out to Muse Ember for parallel execution. No parameter counts, benchmarks, or pricing details were disclosed.
OpenCode announced that Meta's Muse Ember model is now available on its platform, letting developers call the model directly within OpenCode. The announcement provides only availability information, with no details on model parameters, benchmarks, or pricing.
Meta's Muse Ember (Max) tops the Code Arena WebDev small-model leaderboard at 1608 points, beating Claude Haiku 5.5 and GPT-6 Luna. At identical API pricing, its median cost per task is just $0.03, roughly half of Luna and a quarter of Haiku, placing it on the cost-performance Pareto frontier. Generations averaged about two minutes each, producing 20+ websites and animations in a single pass.