Meta Muse Ember Pricing: What $0.10 per 1M Input Tokens Means for Agents
Meta's Muse Ember lists at $0.10 per 1M input tokens, but no output price was published at launch, so the real cost of an agent step depends on tokens per task and the output rate, not the input line alone.
Last updated: November 2026
Key takeaways
- Muse Ember is Meta's fourth Muse model (after Spark, Glimmer and Wonder), built for repeated agent work such as tool calls, routing, extraction and long document parsing.
- Launch coverage gives one number: $0.10 per 1M input tokens. No output price or context window was stated.
- Meta says Ember beats Haiku and Luna on cost per task, using Artificial Analysis's composite score.
- Claude Haiku 5.5 also lists at $0.10 input for prompts up to 100K tokens (per TrueFoundry), so the gap is in output cost and tokens per task, not the input rate.
- Agent workloads are input-heavy, which is where a $0.10 rate matters most.
What does Meta Muse Ember cost?
According to TechCrunch's launch report, Muse Ember costs $0.10 per million input tokens. The article gives no output-token price and no context window. It describes the model as cost-competitive with open models like DeepSeek V4.1 Flash and GLM 5.3 Flash, and says it accepts text, images and audio, so one model can read screenshots, scanned documents and voice memos.
Meta pitches Ember as the worker, with the larger Muse Wonder as the planner. That planner-worker split is the architecture most cost-conscious agent teams already use.
How does Ember compare with Haiku and Luna?
Meta says Ember is cheaper and smarter than both on cost per task. The article does not give Haiku or Luna prices, so here is the closest verified reference, from TrueFoundry's table for Claude Haiku 5.5:
| Model | Input $/1M | Output $/1M | Source | |---|---|---|---| | Muse Ember | $0.10 | not published | TechCrunch | | Claude Haiku 5.5 (prompt up to 100K) | $0.10 | $0.50 | TrueFoundry | | Claude Haiku 5.5 (prompt over 100K) | $0.50 | $2.50 | TrueFoundry |
The interesting row is the second one. On input price alone, Ember and Haiku 5.5 tie below 100K tokens. Any real saving has to come from the output rate, from the long-prompt tier, or from using fewer tokens to finish a task. Treat the cost-per-task claim as a vendor claim until you test it on your own workload. You can line up providers on the Meta pricing page and the Anthropic vs Meta comparison.
Why does input price dominate agent costs?
Agents resend context every step: system prompt, tool definitions, history, retrieved documents. Illustrative example: an agent step with 20,000 input tokens and 1,000 output tokens.
| Ember output rate (illustrative) | Input cost per step | Output cost per step | Cost per 1M steps | |---|---|---|---| | $0.30 per 1M | $0.0020 | $0.0003 | $2,300 | | $0.50 per 1M | $0.0020 | $0.0005 | $2,500 | | $1.00 per 1M | $0.0020 | $0.0010 | $3,000 |
Whatever the output rate turns out to be, input is 67 to 87% of the bill in this shape of workload. That is why a $0.10 input rate is a real lever for agents, and why caching and context trimming matter as much as the model choice. If you are pricing an agent product, start from the AI agent use case calculator.
How should you evaluate a cheap agent model?
- 1Measure tokens per completed task, not per request. Cheap models often need more retries.
- 2Track the cost per successful task: total spend divided by tasks that actually passed your check.
- 3Test a planner plus worker split: a larger model plans, Ember-class models execute.
- 4Wait for the output price before committing a margin to a customer quote.
For a related launch from the same lab, see how we read Meta Muse Spark pricing and contributor discounts.
Frequently asked questions
How much does Meta Muse Ember cost?
Launch coverage lists $0.10 per 1M input tokens. No output price or context window was published in the report we reviewed.
Is Muse Ember cheaper than Claude Haiku?
Meta says yes on cost per task. On input price alone, Haiku 5.5 also lists at $0.10 for prompts up to 100K tokens per TrueFoundry, so the difference would come from output rates and tokens used.
What is Muse Ember designed for?
It targets repeated agent work such as tool calls, coding subtasks, routing, extraction and long document parsing, and it accepts text, images and audio.
Does Muse Ember replace a frontier model?
No. Meta positions it as the high-volume worker that pairs with the larger Muse Wonder planner, so you would keep a bigger model for hard reasoning. The takeaway: judge Ember by cost per successful task, not by the $0.10 headline. Put your own token mix into the Calcaas LLM cost calculator. Note: place the JSON-LD above inside a <script type="application/ld+json"> tag in the page head.
More from the blog
Product UpdatesClaude Fable 5.2 Pricing: $8 Input, $40 Output per 1M Tokens
Claude Fable 5.2 is priced at $8 per 1M input tokens and $40 per 1M output tokens, a 20% cut from Fable 5.1's $10 and $50, with an optional fast mode at $16 and $80 for up to 2.5x faster generation.
Product UpdatesOpenAI DevDay 2026 Recap: What It Means for Your AI Costs
OpenAI DevDay 2026 announced more than 20 products, including GPT-6.1 Sol at $2 input and $10 output per 1M tokens, always-on Dots agents, Ultrafast mode at up to 300 tokens per second and a Pro 500 tier. An always-on agent waking 24 times a day costs roughly $7 to $18 a month.
Product UpdatesClaude Opus 5's Price Didn't Change, But Your Cost-Per-Task Probably Just Dropped
Opus 5 launches at the same $5/$25 per million token price as its predecessor, but Anthropic's own benchmarks suggest it needs fewer tokens and fewer retries to hit the same result, which is the number that actually matters for your margins.
Pricing math, in your inbox.
One short note a week on AI pricing, token economics, and margin. No spam, unsubscribe anytime.