Meta releases Muse Ember, a low-cost model for agent tasks
Article timeline
The articles in this story. Times are ET.
Alexandr WangXVals AI reports Meta Ember is cheapest and fastest in its benchmark top 12AIVals AI reports that Meta's Ember costs $0.15 per test on the Vals Index, the lowest cost of any model on the index. It also completes tests in about 15 minutes, the fastest of the top 12 models, ahead of Haiku 5.5 and GPT-6 Luna.
- The DecoderNewsMeta launches Muse Ember, a small model for routine agent tasks
AIMeta says Muse Ember is its smallest and most efficient AI model, built for tool calls, small coding tasks, routing, and data extraction. Meta claims it offers more intelligence per dollar than Claude Haiku 5.5 and GPT-6 Luna, costing about $0.14 per task on the Artificial Analysis test. It is available now through the Meta Model API and Muse Code at $0.10 per million input tokens and $0.50 per million output tokens.
- Alexandr WangOfficialMeta launches Muse Ember, a fast, low-cost model that matches Haiku 5.5
AIMeta's Alexandr Wang says Muse Ember is fast, fairly smart, and cost effective. Background post from @rynorhn says Muse Ember matches Haiku 5.5 on overall intelligence at about a third of the cost per task, and costs the same per token as GPT-6 Luna while scoring significantly higher.
Vercel DevelopersOfficialVercel adds Meta's Muse Ember model to AI GatewayAIVercel says Muse Ember is now available on AI Gateway. Meta describes it as a lightweight reasoning model for repeatable work such as coding sub-tasks, tool calls, and document summaries.
- Alexandr WangXPickMeta's Muse Ember matches Claude Haiku 5.5 on the Intelligence Index at a third of the cost
AIMeta says its Muse Ember model scores 36 on the Artificial Analysis Intelligence Index, matching Claude Haiku 5.5 (max) at about a third of the cost per task. Artificial Analysis reports Muse Ember (max) at $0.14 per task and says it is the cheapest model scoring 36 or higher. Its AA-Omniscience accuracy of 21.1% is the lowest among the compared similarly intelligent models, though its hallucination rate is 17.4%. Pricing is $0.10 input, $0.004 cached input and $0.50 output per 1M tokens.
- Alexandr WangOfficialMeta releases Muse Ember, its smallest model, at 17x lower price
AIMeta releases Muse Ember, its smallest model yet, which it says offers nearly Muse Spark 1.3 (max) intelligence at a 17x cheaper price point, according to Artificial Analysis. Muse Ember is the first Meta model trained using Muse Wonder Full, and it is available on the API and in Muse Code.
Artificial AnalysisOfficialArtificial Analysis publishes Intelligence Index results for Muse Ember (max)AIArtificial Analysis publishes full Intelligence Index results for Muse Ember at its max setting. The post provides the complete benchmark breakdown but no scores in its text.
- Artificial AnalysisOfficialMuse Ember's AA-Omniscience accuracy trails peers, but its hallucination rate is low
AIArtificial Analysis reports that Muse Ember scores 21.1% accuracy on AA-Omniscience, the lowest among similarly intelligent models it compared. The model declines to answer 64% of questions, which keeps its hallucination rate at 17.4%, half that of Claude Haiku 5.5.
- Artificial AnalysisOfficialMuse Ember leads similarly priced models on the Artificial Analysis Coding Agent Index
AIArtificial Analysis reports that Muse Ember scores 46 on its Coding Agent Index in Muse Code, ahead of Claude Haiku 5.5 in Claude Code at 37 and GPT-6 Luna in Codex at 31. Muse Ember's lead comes from SWE-Atlas-QnA-AA and Terminal-Bench 4.0, while it trails Claude Haiku 5.5 on FrontierCode, as it does on the Intelligence Index.
- Artificial AnalysisOfficialArtificial Analysis reports Muse Ember scores 68.1% on AutomationBench-AA
AIArtificial Analysis reports that Muse Ember scores 68.1% on AutomationBench-AA, level with Muse Wonder and GPT-6 Astra. On agentic coding, Muse Ember beats Claude Haiku 5.5 on Terminal-Bench 4.0 (40.9% vs 32.8%) but trails it on FrontierCode 2.0 (27.2% vs 34.9%).
- Artificial AnalysisOfficialMuse Ember costs less per task than similarly priced models
AIArtificial Analysis reports that Muse Ember (max) has the same input and output token prices as Claude Haiku 5.5 (max) and GPT-6 Luna (max). Its cached input price is 60% lower, and cached input accounts for most of the cost in the group's agentic evaluations. Muse Ember (max) also uses roughly a third as many output tokens as Claude Haiku 5.5 (max).
- Artificial AnalysisOfficialArtificial Analysis finds Meta Muse Ember matches Claude Haiku 5.5 at a third of the cost
AIArtificial Analysis says Meta's Muse Ember scores 36 on its Intelligence Index, matching Claude Haiku 5.5 (max) at $0.14 per task versus $0.41. Muse Ember (max) trails Claude Haiku 5.5 on FrontierCode 2.0 (27.2% vs 34.9%) but beats it on Terminal-Bench 4.0 (40.9% vs 32.8%). Its AA-Omniscience accuracy of 21.1% is the lowest among the compared similarly intelligent models, though its hallucination rate of 17.4% is half of Claude Haiku 5.5's.
- Alexandr WangOfficialMeta announces Muse Ember, a small model it says beats Haiku on cost
AIAlexandr Wang announces Muse Ember, Meta's smallest and most efficient model, built for repeated agent workloads. He says it delivers more intelligence per dollar than Haiku, Luna, and other flash models at any reasoning setting, at a third of the cost or less.
- TechCrunch · AINewsMeta releases Muse Ember, a low-cost model for agent execution
AIMeta Superintelligence Lab released Muse Ember, its lightest and most cost-efficient model yet, designed for tool calls, coding subtasks, routing, extraction and long document parsing. Meta says Ember costs 10 cents per million input tokens, is cheaper than Haiku and Luna on Artificial Analysis's composite benchmark, and can pair with the larger Muse Wonder in agent workflows. It also accepts text, images and audio.
Sophia YangXMeta releases Muse Ember, its smallest model with higher intelligence per dollarAIMeta releases Muse Ember, its smallest model, which it says delivers higher intelligence per dollar than others in its class. The model is built for token-heavy agent work such as tool calls, coding subtasks, routing, extraction, and long document parsing. It is available today on the Meta Model API and in Muse Code.
- Artificial Analysis ArticlesOfficialPickArtificial Analysis finds Meta's Muse Ember cheapest at Intelligence Index score 36
AIArtificial Analysis reports that Meta's Muse Ember (max) scores 36 on its Intelligence Index at $0.14 per task, the lowest cost of any model at that score or higher. It matches Claude Haiku 5.5 (max) on score while costing about a third as much per task. Muse Ember is weakest on knowledge, with 21.1% AA-Omniscience accuracy, because it declines to answer 64% of questions.