Understand this story
Artificial Analysis's evaluation shows that Meta's new model Muse Ember (max) scores 36 on the Intelligence Index, on par with Claude Haiku 5.5 (max), but costs $0.14 per task, about one-third of the latter's. The evaluation also shows that Muse Ember scores 68.1% on AutomationBench-AA, on par with Muse Wonder (max); on Terminal-Bench 4.0 it surpasses Claude Haiku 5.5's 32.8% with 40.9%, but on FrontierCode 2.0 it trails with 27.2% versus 34.9%. Its AA-Omniscience accuracy is only 21.1%, mainly because it refuses to answer 64% of questions, resulting in a hallucination rate of 17.4%. Meta subsequently released Muse Ember on November 10, describing it as its smallest and most efficient model, aimed at complex reasoning execution and everyday agentic work, supporting a 1M token context window and native understanding of text, images, audio and video; it is now available on the Meta Model API, priced at $0.10 per million input tokens and $0.50 per million output tokens.
AI-generated from public reports
Latest development · Report
Meta Releases Its Smallest and Most Efficient Model, Muse EmberDevelopments
3 confirmed developments
Artificial Analysis says Meta's new model Muse Ember scores 36 on the Intelligence Index
Source:Artificial Analysis (full article)- Source:AI at Meta
- Source:AI at Meta
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 10
Simon WillisonSelected
Meta Releases Its Smallest and Most Efficient Model, Muse Ember
Meta AI has released Muse Ember, describing it as its smallest and most efficient model, aimed at complex reasoning execution and tasks that agents perform repeatedly, including tool calling, coding subtasks, routing, extraction, and long-document parsing. Under short prompts, its pricing matches GPT-6 Luna and Claude Haiku 5.5 at $0.10 per million input tokens and $0.50 per million output tokens; beyond each model's respective threshold, prices rise, with Muse Ember at $0.30/$1.50 above 200k. The author tested it across six reasoning levels by generating an SVG of a pelican riding a bicycle, finding that reasoning level does not strictly correspond to the visual quality of the output, and that the high level failed to return a complete SVG.
MarkTechPostRelated discussion
MarkTechPost joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
English text is being prepared.
Meta Launches Lightweight Model Muse Ember, Benchmark Scores Rival Claude Haiku 5.5, Cost Per Task Only 1/3
Meta has launched Muse Ember, its smallest and most efficient model to date, designed specifically for complex reasoning and everyday agent tasks. Meta says it performs comparably to Claude Haiku 5.5 on the Artificial Analysis leaderboard, at only 1/3 the cost per task, and surpasses GPT-6 Luna on multiple tests. The model has a 1 million Token context window, outputs up to 512,000 Tokens, natively supports text, images, audio and video, and is available starting today on Meta Model API, Google Cloud and Oracle Cloud, at $0.10 per million input Tokens and $0.50 per million output Tokens.
IT Home · AIRelated discussion
IT Home · AI joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
SiliconANGLESelected
Meta Releases Entry-Level Model Muse Ember, Focused on Low-Cost AI Agents
Meta has launched its entry-level model Muse Ember, saying it matches Claude Haiku 5.5's score on the Artificial Analysis Index at roughly one-third the price, and surpasses GPT-6 Luna. The model has a 1 million token context window, supports image, audio and video input, and produces outputs of up to 512,000 tokens. It is offered through the Meta Model API at 10 cents per million input tokens and 50 cents per million output tokens, with input dropping to 0.4 cents per million tokens on cache hits. Meta says it is the first model trained using the flagship model Muse Wonder, but did not specify which part of the process was automated.
cnBeta
Meta launches its smallest model Muse Ember, matching Claude's performance at about one-third the cost
Meta has released Muse Ember, its smallest and most efficient model to date, with the model ID muse-ember-1.0, now available through Muse Code and the Meta Model API. The model supports text, image, audio, video, and PDF inputs, outputs text, has a context window of about 1 million tokens, a maximum output of about 512,000 tokens, and charges $0.10 per million input tokens and $0.50 per million output tokens. Artificial Analysis testing shows that at its highest reasoning tier, a single task costs about $0.14, compared with about $0.41 for a similar task on Claude Haiku 5.5; Meta's published tests claim it leads Claude Haiku 5.5 and GPT-6 Luna on some agentic, tool-calling, and multimodal tasks, but scores lower than Claude Haiku 5.5 on FrontierCode 2.0.
cnBetaRelated discussion
cnBeta joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Phoenix Technology
Meta Releases Lightweight Model Muse Ember, Matching Claude Haiku 5.5 in Performance at About One-Third the Cost
Meta released the lightweight model Muse Ember on November 9 local time. Data from third-party evaluation agency Artificial Analysis shows that its composite intelligence index is 36 points, on par with the highest reasoning tier of Claude Haiku 5.5, with a per-task cost of about $0.14, versus about $0.41 for the latter. The model targets high-frequency agent tasks and supports text, image, audio, video, and PDF input, with a context window of about 1 million tokens. It is now available through Muse Code and the Meta Model API under the model ID muse-ember-1.0. Pricing is $0.10 per million input tokens, $0.004 for cached input, and $0.50 for output; its score on the FrontierCode 2.0 test is lower than that of Claude Haiku 5.5.
AI at MetaRelated discussion
AI at Meta joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Rohan Paul
Meta Releases Its Smallest Model Muse Ember, Claims It Matches Claude Haiku 5.5 Intelligence at About One-Third the Cost
Meta released its smallest model, Muse Ember, claiming it reaches Claude Haiku 5.5's intelligence level at about one-third the cost, targeting scenarios such as tool calls, coding subtasks, routing, and extraction that agents repeatedly execute. Pricing is $0.10 per million input tokens and $0.50 per million output tokens for up to 200K tokens, and $0.30 input and $1.50 output beyond 200K tokens. Since Haiku charges the same rate for up to 100K tokens, Ember's cost advantage mainly comes from consuming fewer tokens per task. The model is already available in the Meta Model API and Muse Code.
The Decoder
Meta releases small model Muse Ember, targeting agents' everyday repetitive tasks
Meta has released its smallest and most efficient model, Muse Ember, targeting agents' everyday tasks, including tool calling, small-scale coding, request routing, data extraction, and long-document processing. The model supports text, images, audio, and video, can process up to one million tokens at a time, and is now available through the Meta Model API and Muse Code, priced at $0.10 per million input tokens and $0.50 per million output tokens. Meta says its intelligence per dollar at every reasoning tier is higher than that of Claude Haiku 5.5 and GPT-6 Luna; in Artificial Analysis tests, its cost per task is about $0.14, compared with $2.39 for the larger Muse Spark 1.3, with scores of 36 versus 37; however, on real coding tasks in FrontierCode 2.0-AA, Haiku leads with 34.9% over Muse Ember's 27.2%.
Artificial Analysis (full article)Related discussion
Artificial Analysis (full article) joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
BaoyuRelated discussion
Baoyu joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Unite.AISelected
Meta Releases Its Smallest Model Muse Ember, Targeting Low-Cost Agent and Coding Tasks
Meta AI Research launched Muse Ember on November 9, describing it as the company's smallest model and the best suited for repetitive agent and coding work. It was made available the same day through Muse Code and the Meta Model API, priced at $0.10 per million input tokens and $0.50 per million output tokens. The model is offered as muse-ember-1.0, supports the Responses API, has a 1M-token context window and a maximum output of 512K tokens, and cached input costs $0.004 per million tokens. Meta says it scores 36 on the Artificial Analysis Intelligence Index, below Muse Spark 1.3's 37, but costs $0.14 per task versus $2.39; on FrontierCode 2.0-AA it scores 27.2%, below Claude Haiku 5.5's 34.9%.
The Rundown AI (X)Related discussion
The Rundown AI (X) joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
ArenaRelated discussion
Arena joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Artificial AnalysisSelected
Artificial Analysis evaluates Meta Muse Ember: Intelligence Index score of 36, cost of…
Artificial Analysis's evaluation shows that Meta's new model Muse Ember (max) scores 36 on the Intelligence Index, on par with Claude Haiku 5.5 (max), but costs $0.14 per task, about one-third of the latter's. It is the lowest-cost model among those scoring 36 or above on the index; the xhigh tier scores 34 at $0.12 per task, roughly 4 times cheaper than the equally scoring DeepSeek V4.1 Flash 1017 and GLM-5.3-Flash. Knowledge-related metrics are its weak spot: AA-Omniscience accuracy is 21.1%, and because it chooses not to answer 64% of questions, its hallucination rate drops to 17.4%; in Muse Code it scores 46 on the Coding Agent Index, higher than Claude Haiku 5.5 in Claude Code (37) and GPT-6 Luna in Codex (31). Pricing is $0.10 per 1M input tokens, $0.004 for cached input, and $0.50 for output, and it is now available on Meta's first-party API and in Muse Code.
AI at MetaFirst-partySelected
Meta Releases Lightweight Model Muse Ember, Now Available on Meta Model API
Meta has released its smallest and most efficient model, Muse Ember, now available on the Meta Model API, priced at $0.10 per million input tokens and $0.50 per million output tokens. The model is aimed at complex reasoning execution and everyday agent work: it can carry workflows across applications, call tools in sequence, run tests, and make targeted code fixes. It supports a 1M token context window and native understanding of text, images, audio, and video. Meta says that on the Artificial Analysis Intelligence Index, its performance is comparable to Claude Haiku 5.5, but its cost per task is about one-third of the latter's, and it scores higher than GPT-6 Luna under every reasoning setting.
AI at MetaFirst-partySelected
Meta releases lightweight model Muse Ember, focused on cost-effectiveness for agentic tasks
Meta has released its smallest and most efficient model, Muse Ember, aimed at complex reasoning execution and everyday agentic work; it can take on workflows across applications, call tools in sequence, run tests, and make targeted code fixes. The model supports a 1M-token context window and natively understands text, images, audio, and video. On the Artificial Analysis Intelligence Index, its performance is comparable to Claude Haiku 5.5, with a per-task cost of about one-third of the latter, and it scores higher than GPT-6 Luna across all reasoning settings. Muse Ember is now available on the Meta Model API, priced at $0.10 per million input tokens and $0.50 per million output tokens.
AI at MetaFirst-party
Meta Releases Small Model Muse Ember, Supporting 1M Token Context and Multimodal Input
Meta has released its smallest and most efficient model, Muse Ember, aimed at complex reasoning execution and everyday agent work; it can take on workflows across applications, call tools in sequence, run tests, and make targeted code fixes. The model has a 1M token context window and natively understands text, images, audio, and video. Muse Ember is now available on the Meta Model API, at $0.10 per million input tokens and $0.50 per million output tokens; Meta says it performs comparably to Claude Haiku 5.5 on the Artificial Analysis Intelligence Index, at about one-third the per-task cost of the latter.
Testing Catalog
Meta Releases Small Model Muse Ember, Now Live on Meta Model API and Muse Code
Meta has released its new model Muse Ember, now live on the Meta Model API and in Muse Code. According to Mark Zuckerberg, this is Meta's smallest model, focused on delivering higher intelligence per dollar than models in its class, and aimed at token-intensive work such as agents' all-day repeated tool calls, coding subtasks, routing, information extraction, and long document parsing. TestingCatalog calls it a low-cost small model whose performance is comparable to Claude 5.5 Haiku.
TechCrunch AISelected
Meta releases lightweight model Muse Ember, aimed at agent execution
Meta Superintelligence Lab has released the lightweight model Muse Ember, with an input price of 10 cents per million tokens, targeting high-frequency agent tasks such as tool calling, coding subtasks, routing, extraction, and long-document parsing. Meta says it can be paired with the larger Muse Wonder, with the latter planning and Ember handling high-volume execution, and it supports text, image, and audio input. Ember is the fourth model under the Muse brand, following Spark, the open-source Glimmer, and Wonder.
Artificial AnalysisSelected
Artificial Analysis evaluation: Meta Muse Ember scores 36 on the Intelligence Index, on par with Claude…
Artificial Analysis evaluation shows that Meta's new model Muse Ember (max) scores 36 on the Intelligence Index, on par with Claude Haiku 5.5 (max), but costs $0.14 per task versus $0.41 for the latter. Muse Ember is priced at $0.10 per 1M input tokens, $0.004 for cached input, and $0.50 for output, and is available through Meta's first-party API and Muse Code. Its knowledge-related metrics are relatively weak, with AA-Omniscience accuracy of 21.1%, mainly because it refuses to answer 64% of questions; its hallucination rate is 17.4%, about half that of Claude Haiku 5.5 (max).
Hacker News · dvaplimaRelated discussion
Hacker News · dvaplima joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Nov 9
Artificial Analysis full articleFirst-party
Meta Muse Ember scores 36 on the Intelligence Index, the lowest cost among models with the same score
Artificial Analysis evaluations show that Meta's new model Muse Ember (max) scores 36 on the Intelligence Index, on par with Claude Haiku 5.5 (max), but at a cost per task of $0.14, roughly one-third of the latter's. It scores 68.1% on AutomationBench-AA, on par with Muse Wonder (max); on Terminal-Bench 4.0 it beats Claude Haiku 5.5's 32.8% with 40.9%, but on FrontierCode 2.0 it trails with 27.2% versus 34.9%. Its AA-Omniscience accuracy is only 21.1%, mainly because it refuses to answer 64% of questions, resulting in a hallucination rate of 17.4%. Pricing is $0.10 per 1M input tokens, $0.004 for cached input, and $0.50 for output, and it is available through Meta's first-party API and Muse Code.
Hacker News Trending
Meta releases its smallest and most economical model, Muse Ember, targeting repetitive agentic tasks
Meta has released Muse Ember, its smallest and most efficient model, aimed at high-frequency agentic tasks such as tool calling, coding subtasks, routing, extraction, and long-document parsing. The company says that under equivalent inference settings it is more cost-effective than Claude Haiku 5.5 and GPT-6 Luna, scoring 36 on the Artificial Analysis Intelligence Index at about $0.14 per task. The model supports a 1M token context and up to 512K output tokens, natively understands text, images, audio, and video, and launches today as muse-ember-1.0 on Muse Code and the Meta Model API, priced at $0.10 per million input tokens and $0.50 per million output tokens.
Heat trend for this story
- Current
- 124
- Period peak
- 128 Nov 10 16:00
- 24-hour change
- —
Move the pointer or select the chart to inspect hourly heat. Use the left and right arrow keys with a keyboard.