Competition in AI open source and commercialization continues to intensify. Recently, Meta Superintelligence Labs officially released the fourth model under the Muse brand, and its smallest efficient multimodal reasoning model to date — Muse Ember. This model is built specifically for repetitive subtasks that AI agents need to handle at high frequency during everyday operation.
Core Positioning: Focused on High-Frequency Agent Subtasks, Enabling Fine-Grained Division of Labor
In Meta's product matrix, each model plays a different role. Among them, the flagship Muse Wonder focuses on tackling the most complex planning and hard problems, Muse Spark handles everyday coding and agent workflows, and the new Muse Ember focuses precisely on scaled repetitive coding and agent subtasks.
Meta explicitly states that Muse Ember's core use cases mainly cover tool calls, coding subtasks, routing, extraction, and long document parsing. Through this fine-grained division of labor, large models can support the daily operation of massive agents with greater efficiency.
Architectural Features: Native Multimodal Fusion and a Million-Token Context Window
In terms of underlying architecture and multimodal capabilities, Muse Ember demonstrates a very strong overall processing level. The model natively supports multiple input forms, including text, images, audio, and video. This means that when agents perform tasks, they do not need to call separate models; they can directly read screenshots, charts, scanned pages, voice memos, or even meeting recordings, and can even operate desktop applications directly through screenshots.
In terms of processing capacity, Muse Ember has an ultra-large context window of up to 1 million tokens, and a single call can easily accommodate a complete codebase or an entire document collection. At the same time, its maximum output length can reach 512,000 tokens, four times that of similar competitors.
It should be noted that the model uses a fully managed model, running only on Meta's managed infrastructure (namely Meta Model API and Muse Code). Its weights remain private, so it does not offer options for running on local GPUs, laptops, or quantized deployment.
Benchmarks: Stunning Performance on Agent Tasks, Slightly Behind in Code Merge Scenarios
In evaluations by authoritative testing organizations and Meta's internal test suites, with maximum inference effort enabled, Muse Ember demonstrated excellent overall performance:
Agents and workflows (Agents): In the AutomationBench-AA test based on 657 Zapier business workflows, Ember's task completion rate reached 37.8%, far ahead of GPT-6Luna's 20.2% and Claude Haiku5.5's 13.9%. In the internal Agents’ Last Exam test, Ember scored 61.6, also better than Haiku's 57.0 and Luna's 57.1.
Coding and development (Coding): In the Terminal-Bench4.0 test, Ember led with 40.9 points, ahead of Haiku's 39.2 and Luna's 16.4; it also ranked first with 46 points in the Artificial Analysis (AA) coding agent index. However, in the FrontierCode2.0-AA test, which examines code merge capability in real repositories, Haiku5.5 scored 34.9, outperforming Ember's 27.2.
Multimodal and perception: In the GDP.pdf-AA test, Ember tied with GPT-6Luna (22.8 points); on OSWorld2.1 (partial score), it held a narrow lead over Haiku with 69.5 points versus 68.9; in the tool-integrated PerceptionBench test, Ember scored 80.6, higher than Haiku's 78.6 and Luna's 62.8.
In the overall Artificial Analysis Intelligence Index, Muse Ember scored 36, exactly on par with Claude Haiku5.5 (max effort), and ahead of GPT-6Luna's 28.
Cost Efficiency: Same Base Pricing, Significantly Optimized Actual Task Costs
In terms of public API pricing, Muse Ember's standard rates are consistent with competitors: $0.10 per million input tokens, $0.50 per million output tokens, and $0.004 per million cached input tokens (lower than Haiku and Luna's $0.01).
Although the base unit price is similar, thanks to improved token consumption efficiency, the actual cost of executing a single task has been significantly optimized. Test data shows that Muse Ember's cost per task under the highest reasoning setting is about $0.136, while Claude Haiku5.5 costs as much as $0.411 to complete similar tasks, meaning its cost per task has dropped to about one-third that of its competitor.
In addition, when handling ultra-long text, competitors often double their prices or apply tiered surcharges once prompts exceed 10K or 272K, while Meta has not set up a separate long-context surcharge tier for Ember, further strengthening its cost-effectiveness in long-text processing.
Ecosystem Access and Partners
Currently, developers can officially access and use Muse Ember through Meta Model API and Meta's coding agent tool Muse Code, and it supports the standard Responses API. During the early testing phase, many well-known companies and technology platforms, including Box, Cline, Oracle, Cloudflare, and You.com, have already participated.
