Muse Ember 1.0 (Meta)

Ember is the small model in Meta's Muse lineup, with a 1 million token context window, native support for text, images, audio, and video, a 512,000-token maximum output on the direct Meta API, and the Responses API model ID muse-ember-1.0. It is also listed in Muse Code, and the article notes that third-party gateway endpoints may accept fewer input modalities than Meta's native API. On Meta's launch chart for Artificial Analysis's Intelligence Index, Ember scores 35.8 points at $0.136 per benchmark task, versus 36.2 and $0.411 for Claude Haiku 5.5 at Max. On FrontierCode 2.0, which judges whether engineering work is mergeable, Haiku leads Ember by 7.7 percentage points, so the cost advantage does not extend to mergeable-code quality.

Meta released Muse Ember on November 9, 2026, as a lightweight model built for agent execution, with the model ID muse-ember-1.0. It carries a 1 million token context window and a 512,000 token output ceiling, and Meta prices it at $0.10 per million input tokens and $0.50 per million output tokens. Developers can call it through Muse Code and the Meta Model API. Meta positions Muse Ember as its smallest and most efficient model, designed for complex reasoning execution rather than planning. Meta's developer documentation, published as a Muse Ember cookbook, names sub-agent tasks and processing collections of images, files, and records as target uses, covering tool calls, coding subtasks, routing, extraction, and long-document parsing.

Muse Ember, Meta's smallest and cheapest model released November 9, 2026, lists at $0.10 per million input tokens and $0.50 per million output tokens, the same sticker price as Claude Haiku 5.5 and GPT-6 Luna, and targets the small repeated steps agents run all day such as tool calls, routing, extraction, and quick code fixes. Its difference is token efficiency rather than per-token price. On Artificial Analysis's Intelligence Index, Ember scores 36 for $0.14 per task while Haiku 5.5 scores 36 for $0.41, because Haiku wrote 298M output tokens against Ember's 115M. Ember wins 7 of Meta's 10 comparison rows, ties 2, and loses 1, with Claude Haiku 5.5 ahead on FrontierCode 2.0 at 34.9 versus 27.2, and prompts over 200,000 tokens priced three times higher at $0.30 in and $1.50 out per million tokens.

BenchLM's catalog entry for Muse Ember lists a 1M token context, $0.10 input, $0.50 output, and $0.004 cached input pricing, with a $0.30 blended rate, and records it among 214 ranked models. The tracker is explicit about evidentiary limits: only four published benchmark rows are displayable, category scores and comparative ranks remain pending, and independent runtime speed plus coding, reasoning, knowledge, multilingual, math, and instruction-following categories are not measured. Because no category has enough eligible evidence for a comparative rank, the entry presents capability as a field-median comparison rather than an absolute quality threshold. The page offers its rankings, source results, and catalogs through an API or MCP for programmatic access, and flags that independent runtime speed has not been measured for this model.

Meta launched Muse Ember as a lightweight model for repetitive tasks inside agent workflows, including tool calls, coding subtasks, request classification, data extraction, and reading lengthy documents. Available since November 9, 2026 in Muse Code and the Meta Model API, it starts at $0.10 per million input tokens and is presented as Meta Superintelligence Labs' smallest and most efficient model to date. The model accepts text, images, audio, video, and PDFs but generates text, with a context window of 1,048,576 tokens and up to 524,288 output tokens. Meta separates planning from execution: Muse Wonder or Muse Spark plan while Ember runs delegated coding tasks, terminal commands, chained tool calls, routing, and batch processing of images, files, and records, with support for computer use, structured output, and search grounding.

Meta unveiled Muse Ember on November 9, 2026, released as muse-ember-1.0 and available through the Muse Code platform and the Meta Model API at $0.10 per million input tokens, $0.004 per million cached input tokens, and $0.50 per million output tokens. It offers a 1-million-token context window with up to 512K output tokens, and Meta says it costs roughly $0.14 per Artificial Analysis benchmark task. Partner testing highlighted latency and efficiency: Cloudflare's CTO Dane Knecht reported Ember completed every coding test at the lowest measured latency with a p95 of 2.2 seconds, roughly 4x to 13x faster than leading open-weight models while using 36-48% fewer output tokens. Oracle reported up to 30% faster tail latency, and Box praised accuracy on document-heavy tasks. Meta also says Muse Ember is the first model trained using Muse Wonder Full.