Meta releases Muse Ember, claiming Haiku-level intelligence at a third of the cost
Meta released Muse Ember, its smallest and most efficient AI model, built for token-heavy agent workloads such as tool calls, coding subtasks, routing, extraction and long document parsing. Meta claims it matches Claude Haiku 5.5's top Artificial Analysis Intelligence Index score of 36 at roughly one-third the cost per task, and beats GPT-6 Luna. It is available on the Meta Model API and Muse Code at $0.10 per million input tokens and $0.50 per million output tokens, with a 1M-token context window and multimodal support.
Reference imageEditorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary covers the current reports
Cross-source coverage
Reporting timeline
Meta's Muse Ember beats Claude Haiku 5.5 and GPT-6 Luna on agent and coding benchmarks
Meta released Muse Ember, its first model trained using Muse Wonder Full, which the post says beats Claude Haiku 5.5 and GPT-6 Luna on key agent and coding benchmarks. According to the post, Ember matches Haiku's overall intelligence score at roughly one-third the cost per task, with all three models starting at $0.10/$0.50 per million input/output tokens. Ember supports text, images, audio and video. The post characterizes the launch positively, calling it a great release and saying Meta is "super back."
Read sourceMeta's Muse Ember scores 36 on Intelligence Index at a third of Claude Haiku 5.5's cost
Meta released Muse Ember, a new efficient-tier model line, which scores 36 on the Artificial Analysis Intelligence Index, matching Claude Haiku 5.5 (max) at $0.14 per task versus $0.41 for Claude Haiku 5.5. Muse Ember (xhigh) scores 34 for $0.12, and the model sits on the Intelligence Index vs. Cost per Task Pareto frontier at both max and xhigh effort. It trails Muse Wonder (max, 44, $6.95) by 8 points at roughly 1/50th the cost per task. On AutomationBench-AA it scores 68.1%, level with Muse Wonder and ahead of Claude Haiku 5.5 on GDP.pdf, though behind on AA-Briefcase. In agentic coding it beats Claude Haiku 5.5 on Terminal-Bench 4.0 but trails on FrontierCode 2.0, and in Muse Code it scores 46 on the Coding Agent Index. Its main weakness is knowledge, with 21.1% AA-Omniscience accuracy, largely because it declines to answer 64% of questions, keeping hallucination at 17.4%. Pricing is $0.10 input / $0.004 cached input / $0.50 output per 1M tokens, available via Meta's first-party API and Muse Code.
Read sourceMeta Introduces Muse Ember, Its Smallest and Most Efficient AI Model
Meta announced Muse Ember, described as its smallest and most efficient model, built for complex reasoning execution and everyday agent work. According to the post, Muse Ember can carry workflows across apps, call tools in sequence, run tests, and make targeted code fixes. It features a 1M-token context window and native understanding of text, images, audio, and video, allowing it to work across large document sets and varied inputs. The announcement positions the model for agentic tasks and multimodal document handling, though no benchmark results, pricing, availability date, or licensing details were disclosed in the post.
Show 3 older updatesHide older updates
Meta Releases Muse Ember, a Small Efficient Model for Agent Workloads
Meta announced Muse Ember, described as its smallest and most efficient model, aimed at complex reasoning execution and repetitive agent tasks such as tool calls, coding subtasks, routing, extraction, and long document parsing. The company claims Muse Ember offers more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna at any reasoning setting, matching Haiku's best score at roughly a third of the cost and beating every Luna setting, and says it sits on the efficiency frontier of the Artificial Analysis leaderboard. Cited benchmark results include AutomationBench-AA 37.8, Terminal-Bench 4.0 40.9, and a General Artificial Analysis Index of 36. Muse Ember is available on the Meta Model API at $0.10 per million input tokens and $0.50 per million output tokens, with costs described as estimates at proposed pricing. Meta says early partners used it for tasks like contract review and code writing.
Read sourceMeta releases Muse Ember 1.0 at $0.14 per task, 17× cheaper than Muse Spark 1.3
Meta has released Muse Ember 1.0, a new model priced at $0.14 per task, according to a post on X. The post compares it with Muse Spark 1.3, which it says costs $2.39 per task, describing Ember as 17× cheaper for just one point less. The post also states that Ember is the first model Meta trained using Muse Wonder Full. No further details on benchmark scores, task definitions, or availability were provided in the source item.
Read sourceMeta Releases Muse Ember, Its Smallest Model, for Token-Heavy Agent Work
Meta announced the release of Muse Ember, described as its smallest model and positioned as delivering higher intelligence per dollar than others in its class. The model is built for token-heavy work that agents repeat throughout the day, including tool calls, coding subtasks, routing, extraction, and long document parsing. Muse Ember is available today on the Meta Model API and in Muse Code. The post frames the release around cost efficiency for high-volume agent workloads rather than frontier capability.
Read source