---
type: "post"
id: "meta-released-muse-ember-its-smallest-reasoning-model-98fb00"
slug: "meta-released-muse-ember-its-smallest-reasoning-model-98fb00"
title: "Meta released Muse Ember, its smallest reasoning model, matching Claude Haiku 5.5 intelligence at one-third the cost."
description: "Artificial Analysis lists Meta’s Muse Ember (Max) as a proprietary reasoning model with text and image input, a one-million-token context window and API pricing of $0.10 per million input tokens and $0.50 per million output tokens."
retrieval_nugget: "Artificial Analysis lists Meta’s Muse Ember (Max) as a proprietary reasoning model with text and image input, a one-million-token context window and API pricing of $0.10 per million input tokens and $0.50 per million output tokens. Its benchmark results make it a candidate for teams to test, but do not establish how it will perform in their own workflows."
published_at: "2026-11-10"
updated_at: "2026-11-10"
record_date: "2026-11-10"
date_kind: "published_at"
topics: ["claude"]
entities: ["Meta","Muse Ember","Artificial Analysis","Claude Haiku 5.5","GPT-6 Luna"]
related_patterns: []
novelty_label: "new_fact"
what_changed: "Artificial Analysis ranks Muse Ember No. 2 of 33 on its Intelligence Index and reports $0.14 average cost per evaluation task."
prior_refs: ["https://newruntime.com/posts/meta-launched-muse-ember-its-smallest-model-built-4baf24/"]
source_url: "https://artificialanalysis.ai/models/muse-ember"
source_urls: ["https://artificialanalysis.ai/models/muse-ember"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:meta-released-muse-ember-its-smallest-reasoning-model-98fb00"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/meta-released-muse-ember-its-smallest-reasoning-model-98fb00/","markdown":"https://newruntime.com/posts/meta-released-muse-ember-its-smallest-reasoning-model-98fb00.md","json":"https://newruntime.com/posts/meta-released-muse-ember-its-smallest-reasoning-model-98fb00.json"}
markdown_url: "https://newruntime.com/posts/meta-released-muse-ember-its-smallest-reasoning-model-98fb00.md"
---

# Meta released Muse Ember, its smallest reasoning model, matching Claude Haiku 5.5 intelligence at one-third the cost.

## Retrieval answer

Artificial Analysis lists Meta’s Muse Ember (Max) as a proprietary reasoning model with text and image input, a one-million-token context window and API pricing of $0.10 per million input tokens and $0.50 per million output tokens. Its benchmark results make it a candidate for teams to test, but do not establish how it will perform in their own workflows.

## Summary
Artificial Analysis lists Meta’s Muse Ember (Max) as a proprietary reasoning model with text and image input, a one-million-token context window and API pricing of $0.10 per million input tokens and $0.50 per million output tokens. Its benchmark results make it a candidate for teams to test, but do not establish how it will perform in their own workflows.

## What happened
Artificial Analysis lists Muse Ember (Max) as a Meta model released in November 2026, with a specific release date of November 9 in its FAQ. The page labels it proprietary and identifies it as the reasoning version; it notes that a non-reasoning variant may also exist. The model accepts text and image input and produces text output. Its listed context window is one million tokens. Artificial Analysis says the model is available through one API provider and that there is no publicly available API on the page’s benchmark comparison views. That distinction matters: the listing reports provider access, but not a generally available Meta endpoint or a choice of multiple providers.

The page reports an Artificial Analysis Intelligence Index score of 36, ranking it second among 33 models in the displayed comparison. The comparable-model median is 24. The v5.0.1 index incorporates nine evaluations, including agentic knowledge work, coding, science, professional document reasoning and general knowledge. Artificial Analysis reports that completing an index task took 3.2 minutes on average, compared with a 7.5-minute median for comparable models. Its reported time to first answer token is 4.74 seconds. The site says speed and latency figures average endpoint runs over the preceding five days and that it runs the benchmark every six hours.

Artificial Analysis lists API prices of $0.10 per million input tokens and $0.50 per million output tokens, based on Meta’s API pricing. It also reports a 96% cache discount and gives a blended rate of $0.0728 per million tokens using a 7:2:1 cache-hit/input/output ratio; it cautions that provider pricing may vary. On its own Intelligence Index workload, the site calculates an average cost of $0.14 per task. That is a benchmark-specific weighted estimate, not a general price per agent task.

The model generated 115 million output tokens across the index evaluation, compared with a median of 90 million for models in the same price tier. Artificial Analysis characterizes it as somewhat verbose. The listing says model weights are not publicly available and Meta has not disclosed the parameter count. Its figures therefore describe a closed model accessed through a provider, not a downloadable model that teams can run or inspect themselves.

## What the archive adds
New Runtime’s earlier coverage on [November 9](https://newruntime.com/posts/meta-launched-muse-ember-its-smallest-model-built-4baf24/) described Meta’s Muse Ember as a small model for agent subtasks and tool calls, with cost-performance claims against Claude Haiku 5.5 and GPT-6 Luna. The Artificial Analysis listing adds benchmark, speed, context-window and API pricing figures; it is a new measurement of a known story, not a separate release.

## Trend signal
This supports, but does not prove, the idea that model comparisons are becoming more useful when expressed as task-level cost and time rather than token prices alone. Artificial Analysis gives a benchmark-specific cost per index task, but the index task is not necessarily a verified completed task in a production workflow. This does not yet fit a pattern New Runtime tracks.

## Why it matters
A low token price can make a model worth testing for high-volume work, but it is not enough to decide whether to route production traffic to it. The useful comparison is the cost of completing a team’s actual task at an acceptable quality level, including retries, long reasoning outputs, cache behavior and human review. Artificial Analysis’s figures offer a starting point: the model’s listed prices are low relative to the page’s comparison medians, and its benchmark task time is also below that median. Yet the same evaluation shows higher-than-median output volume, which can affect both cost and latency in real deployments.

The listing also puts practical constraints in view. A one-million-token context window may matter for document-heavy work, while image input creates a possibility for multimodal tasks. But the page names only one API provider and the model weights are unavailable. Teams should verify access, provider-specific billing and operational limits before designing around either capability. Treat the benchmark as a shortlist signal, then compare Muse Ember with the current model on representative tasks and measure successful completion, error handling and total cost.

## Limits and uncertainty
Artificial Analysis’s index combines nine evaluations; its score is not a direct measure of performance in a particular company’s workflow. The reported $0.14 cost applies to its weighted index tasks, and the stated token prices may differ by provider. The listing identifies one API provider but does not establish broad availability or disclose operational quotas. Meta has not disclosed the model’s parameter count, and the proprietary weights prevent independent inspection or self-hosting. The benchmark figures are reported by Artificial Analysis and have not been independently checked here.

## Reader takeaway
Muse Ember is worth a controlled comparison if low-cost reasoning, long context or image input matters to your workload. Its published benchmark and price figures are useful screening data, not a substitute for measuring the cost and quality of your own completed tasks.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Claude](https://newruntime.com/topics/claude.md): Explore the claude topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)