Meta Superintelligence Lab released Muse Ember on Monday, positioning it as the lab’s lightest and most cost-efficient artificial intelligence model to date. According to TechCrunch AI, the model is priced at 10 cents per million input tokens, a figure that undercuts frontier competitors such as Anthropic’s Haiku and OpenAI’s Luna.reportedMeta launched a new, affordable AI model called Ember that helps computers handle repetitive tasks like reading documents and making tool calls.Meta Superintelligence Lab introduced Muse Ember, a lightweight model priced at 10 cents per million input tokens for high-throughput agent execution.
Evidence: reported
Meta releases Muse Ember, a new lightweight model for agent executionaifleet-techcrunch
The figures
The price point of 10 cents per million input tokens is the central metric according to the published report This cost applies to input tokens processed by the Muse Ember model.reported TechCrunch AI notes that this price is significantly cheaper than the rates for Haiku and Luna. Additionally, the report states that Ember’s cost structure is competitive with open-source alternatives like DeepSeek V4.1 Flash and GLM 5.3 Flash.reported
Evidence: reported
Meta releases Muse Ember, a new lightweight model for agent executionaifleet-techcrunch
Evidence: reported
Meta releases Muse Ember, a new lightweight model for agent executionaifleet-techcrunch
Behind the figures
Meta used Artificial Analysis’s composite benchmark score to demonstrate that Ember is both cheaper and smarter than its competitors. TechCrunch AI reports that Meta emphasized cost-per-task in its release materials.reported The company stated that Ember is designed for complex reasoning execution and repetitive agent work, including tool calls, coding subtasks, routing, extraction, and long document parsing.
Evidence: reported
Meta releases Muse Ember, a new lightweight model for agent executionaifleet-techcrunch
What it means for your stack
Meta designed Ember to pair with larger models, such as Muse Wonder, within an agentic harness. TechCrunch AI explains that a heavier model might establish rules for handling issues, while Ember applies those rules across tens of thousands of cases. Meta described this dynamic as “Muse Wonder plans, Muse Ember handles the volume.” This architecture allows developers to offload high-throughput tasks to the cheaper model.
Building on this
Developers can use Ember to process diverse media types without deploying separate specialized models. According to the release, agents can read screenshots, charts, and scanned documents, or work from voice memos, meeting recordings, and video.reported This multimodal capability allows inference engineers to build agents that ingest mixed input formats directly. Ember is the fourth model under the Muse brand, following Spark, Glimmer, and Wonder, providing a tiered option for different computational needs.
Why it matters
- Inference engineers can reduce compute costs for high-volume agent tasks by using Ember’s 10-cent-per-million-token pricing.
- The model’s multimodal training allows developers to handle text, images, and audio inputs within a single agent workflow.
- Pairing Ember with heavier models like Muse Wonder enables a split architecture for planning versus execution at scale.
- Ember’s cost competitiveness with open-source models like DeepSeek V4.1 Flash offers a viable alternative for budget-conscious deployments.
Topics: Meta, AI models, agent execution, inference costs, multimodal AI
Updated
Then / now / next
Meta’s Muse portfolio included mid-weight Spark, open-source Glimmer, and frontier Wonder models.
Meta Superintelligence Lab added Muse Ember as a lightweight, cost-efficient option for agent execution.
Developers may test Ember’s performance in agentic harnesses paired with heavier planning models.
Story timeline
- Then Meta’s Muse portfolio included mid-weight Spark, open-source Glimmer, and frontier Wonder models.
- 09 Nov 2026 Meta releases Muse Ember, a new lightweight model for agent execution
- Now Meta Superintelligence Lab added Muse Ember as a lightweight, cost-efficient option for agent execution.
- Next Developers may test Ember’s performance in agentic harnesses paired with heavier planning models.
The strongest doubt
Counterpoint from the sources
The source material does not provide independent benchmarks or third-party cost comparisons beyond Artificial Analysis’s composite score cited by Meta.
Ask the article
Local keyword search of this article; no generated answers.Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Terms in this story
input tokens
Units of text or data processed by an AI model during inference.
agentic harness
A software framework that orchestrates AI models to perform complex, multi-step tasks.
multimodal
Capable of processing multiple types of data, such as text, images, and audio.
Questions readers ask
How much does Muse Ember cost per million input tokens?
Meta prices Muse Ember at 10 cents per million input tokens, according to the published report
Which models is Ember cheaper than?
TechCrunch AI reports that Ember is significantly cheaper than Anthropic’s Haiku and OpenAI’s Luna.
Can Ember process audio and video inputs?
Yes, Meta stated that Ember can work from voice memos, meeting recordings, and video without a separate model.
Connected stories
Reading receipt
- Meta Superintelligence Lab released Muse Ember on Monday as its lightest and most cost-efficient AI model.
- The model costs 10 cents per million input tokens, which is lower than Anthropic’s Haiku and OpenAI’s Luna.
- Ember is designed to handle volume-heavy tasks like tool calls and document parsing alongside heavier models.
Still open: How does Ember’s accuracy compare to Haiku and Luna on specific coding subtasks?
Still open: What are the latency implications of using Ember for real-time voice memo processing?