Short answer
Meta released Muse Ember on November 9, 2026, describing it as its smallest and most cost-efficient model yet. The model ID is muse-ember. In the official materials the name appears as meta/muse-ember, where meta/ is the vendor namespace that AI Gateway adds — it is not a second half of the model name.
The facts, all from this launch cycle:
- Role: a lightweight reasoning model for the high-volume, bounded parts of an agent run.
- Context: 1 million tokens.
- Input: text, images, video.
- Tool calling: yes — each result can feed the next step.
- Reasoning effort: adjustable from
minimaltomax. - Price: Meta has not published an API rate.
muse-ember is not on TeamoRouter yet. Once it is live, the key and base URL you already use stay put; only model changes to muse-ember. There is no Meta model on TeamoRouter today, so this article does not name a stand-in — there is nothing from the same vendor in the catalog to point at.
What this model is for
Meta's framing: Muse Ember is the company's smallest and cheapest model, and it delivers more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna. That is a vendor claim, and this launch cycle ships no benchmarks and no unit price to back it, so treat it as direction rather than a measured result.
Vercel's same-day integration note is more concrete about the job. Muse Ember handles the high-volume parts of an agent run — coding sub-tasks, tool calls, document analysis. In other words, the bounded work an agent repeats many times inside a single run: implementing a helper function, calling a tool, summarizing a document.
The 1M-token window is built for exactly that. Code, screenshots, documents and clips can go into the same run without constant trimming and restarting.
What Meta announced
The details below come from Vercel's integration note and the TechCrunch report from November 10.
- Multimodal input: text, image and video share one 1M-token window.
- Tool calling: your application can pass each tool result into the next step.
- Adjustable reasoning effort:
minimalthroughmax, set per task. - Several call surfaces: AI SDK, Chat Completions, Responses and Anthropic Messages, plus coding agents.
- Size and cost: Meta describes it as its smallest and most cost-efficient model.
One thing worth stating up front: this cycle does not list what the model cannot do. Image generation, audio, caching, batch — none of it appears in either source, so this article does not fill the gap for Meta. Check the official model card before you write any of that into an architecture doc.
Pricing
If you are searching for Muse Ember pricing, here is the blunt answer: Meta has not published an API rate.
Only two statements are citable:
- Meta says Muse Ember delivers more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna. That is a relative claim, not a price list, and no numbers accompany it.
- Vercel says AI Gateway passes provider pricing through with no markup and charges no platform fee on inference, including on Bring Your Own Key requests. That describes how the route is billed, not what Muse Ember itself costs.
There is no official figure for input or output cost per million tokens to quote here. If you see a specific rate, check whether it came from Meta's developer pricing page — do not take a third-party paraphrase as the official price.
How the official docs say to call it
Meta's documented entry point is AI Gateway, with the model name written meta/muse-ember. With the AI SDK:
import { streamText } from 'ai';
const result = streamText({
model: 'meta/muse-ember',
prompt: 'Write a JavaScript helper that deduplicates strings while preserving order.',
});
The same note lists Chat Completions, Responses and Anthropic Messages as alternative formats, plus coding agents such as Claude Code, Codex and Cursor connected to AI Gateway. The Anthropic Messages shape looks roughly like this (point the URL at whatever endpoint you actually use):
curl https://<your-endpoint>/v1/messages \
-H "x-api-key: YOUR_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "Content-Type: application/json" \
-d '{"model": "meta/muse-ember", "max_tokens": 1024, "messages": [{"role": "user", "content": "Explain what a context window is in one sentence."}]}'
One caveat: neither source gives Meta's own native API format. What Meta lists is the compatibility set above, so this article does not invent a "native Meta endpoint" to fill the slot.
Calling it on TeamoRouter (once it lands)
Not available yet. When muse-ember enters the catalog, your key and base URL stay as they are — only model changes:
from openai import OpenAI
client = OpenAI(base_url="https://api.teamorouter.com/v1", api_key="YOUR_KEY")
resp = client.chat.completions.create(
model="muse-ember",
messages=[{"role": "user", "content": "Write a JS function that deduplicates an array of strings while preserving order."}],
)
print(resp.choices[0].message.content)
Leave reasoning effort and tool definitions until the model is listed — use the parameter names on TeamoRouter's model page rather than copying field names from the gateway examples.
muse-ember is not on TeamoRouter yet. Once it is live, the same key and model: muse-ember is all you change; there is no Meta model on the catalog today, so there is no substitute to recommend.
What to use today
This article names no stand-in, for a simple reason: TeamoRouter has no Meta model right now. A lightweight model from another vendor will differ in context length, tool-calling conventions and reasoning-effort defaults, so an agent tuned for Muse Ember will likely need re-tuning. That is not a model-name swap.
The realistic move is to wait for it to land. If you want to start writing code now, put the model name behind a config value and change one line on launch day.
How to pay
No subscription: top up your balance in the console from $5 and each request is billed by the tokens it uses. Cards are supported, and so are USDT, USDC and ETH. The same key also calls Claude and GPT models.
New accounts start with a $0 balance — top up before your first request. Top up from $5. If a request returns 402 (400 on the OpenAI format) with "wallet balance is insufficient", your balance is empty — your base URL and key are fine. Top up and retry.
FAQ
Is Muse Ember out?
Yes. Meta released it on November 9, 2026. Vercel's note the same day announced meta/muse-ember on AI Gateway, and TechCrunch covered it on November 10.
Is the model ID muse-ember or meta/muse-ember?
The model ID is muse-ember. The official materials write meta/muse-ember, where meta/ is the vendor prefix AI Gateway puts in front so one API can address many sources.
How much does it cost?
Meta has not published an API rate. The only citable statements are the relative "more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna" claim and Vercel's note that AI Gateway passes provider pricing through without markup. There is no number that belongs in a cost table yet.
How long is the context window?
1 million tokens, shared across text, image and video input.
Does it support tool calling?
Yes. Tool results can be fed into the next step, which is what makes it usable for an agent chaining many small tasks.
Can I adjust reasoning effort?
Yes, from minimal to max. Use the low end for simple helper functions and move up for harder sub-tasks.
Can it generate images?
This cycle does not mention image generation, so do not assume it. Wait for the official model card.
Can I call muse-ember on TeamoRouter now?
No. Meta shipped it, but it has not entered our catalog. Once it does, the same key and base URL, with only model changed.
Once it is live, is it just a model-name swap?
Yes. The documented usage is a model name — meta/muse-ember — with no separate endpoint behind it, so on TeamoRouter it is the same story: change model.
Related reading
- How to get access to Muse Ember: the routes available today
- Muse Ember vs Claude Haiku 5.5 vs GPT-6 Luna: the full comparison
- Vercel AI Gateway announcement: the official source