Meta has a developer stack now. Muse Spark on the Meta Model API, a terminal agent called Muse Code, an image model at a cent a picture, and transcription by the hour. Here is what each one is, from Meta’s own developer site — including the token price, which this site could not find on 16 Sep 2026 and which Meta lists as of 22 Sep 2026.Meta, Products & Solutions for AI Developers, read in a browser 16 Sep 2026; the site is script-rendered and returns nothing to an automated fetch.First-hand: until 22 Sep 2026 this sentence ended “and the one number it will not publish”. The price and its source are in the section below.
Four products, each now with a price from Meta. The text model’s price was the gap on 16 Sep 2026; by 22 Sep 2026 it was listed.
- Muse Spark 1.3 is the model, reached through the Meta Model API.
- Muse Code is a terminal agent, installed with one command. It came out of beta on 31 Aug 2026.Meta, Muse Code: New plans and features, Aug 31, 2026, read at source 22 Sep 2026: “Muse Code is now out of beta with new features and updates designed to help you work faster and smarter.”
- Muse Image costs $0.01 per image — the lowest of the three published image prices compared on this page, against Google and xAI.Meta, AI developer site, read in a browser 16 Sep 2026: “Muse Image is now on Meta Model API at $0.01/image” and “Leading streaming transcription at $0.18/hour”.
- Muse Voice Transcribe is $0.18 an hour.Meta, AI developer site, read in a browser 16 Sep 2026: “Muse Image is now on Meta Model API at $0.01/image” and “Leading streaming transcription at $0.18/hour”.
- Muse Spark 1.3 is $1.25 per million input tokens and $4.25 per million output, or $0.10 and $0.20 on a contributor tier Meta labels “Used to improve our products.” A 22 Sep 2026 snapshot; re-check before budgeting.Meta, Meta Model API, read at source 22 Sep 2026: the table rows read “muse-spark-1.3 Not used to improve our products. 1M $1.25 $0.15 $4.25” and “muse-spark-1.3-contributor Used to improve our products. 1M $0.10 $0.002 $0.20” (context window, then input, cached input and output per million tokens). The same table is on Meta’s Muse Spark 1.3 page, read the same day. Until 22 Sep 2026 this line said Muse Spark’s token price was not published anywhere on the site.
Every price here was correct when this site last checked it, between 16 Sep and 22 Sep 2026. Prices change often and this site no longer updates them, so check the vendor’s own page before you rely on one: Meta pricing · Google pricing · xAI pricing.
The four parts
| Product | What Meta calls it |
|---|---|
| Meta Model API | “Direct, self-serve and affordable access to Muse Spark” |
| Muse Code | “Multi-agent coding in your terminal” |
| Muse Image | “Agentic image gen”, $0.01/image |
| Muse Voice Transcribe | “Leading streaming transcription at $0.18/hour” |
Meta, Products & Solutions for AI Developers, read in a browser 16 Sep 2026.
On 16 Sep 2026 Meta published a price for the image model and the transcriber but not for the text model, the one most people would be budgeting for. By 22 Sep 2026 the text model had a price too — and a much cheaper version that Meta says is used to improve its products.First-hand: until 22 Sep 2026 this takeaway said Meta did not publish a price for the text model.
A cent an image
Meta describes Muse Image as “A reasoning model that plans layout before drawing, with anchored composition for consistent edits”, and gives the price outright: “Muse Image is now on Meta Model API at $0.01/image.”
For comparison, Google’s Nano Banana 2 works out around 6.7 cents for a 1K image and xAI charges $0.04 on grok-imagine-image-2.0. If Meta’s figure holds under real use, it is several times cheaper than either.Google, Gemini Developer API pricing, read in a browser 16 Sep 2026: 1K images “consume 1120 tokens and are equivalent to $0.067 per image” · xAI, Pricing, read at source the same day: grok-imagine-image-2.0 “$0.04 / image”.
grok-imagine-image-2.0 “$0.04 / image”.Cheapest published is not cheapest in practice. An image model that needs three attempts at a cent each is not cheaper than one that works first time at four. Nothing here was generated or compared.
The price that was not there
When this page was first checked, on 16 Sep 2026, the Meta Model API was described as “affordable” without a number attached, and this site found no pricing page on the developer site. By 22 Sep 2026 Meta’s Model API page and its Muse Spark 1.3 page both carried a price table. This site does not know the date it went up.First-hand: until 22 Sep 2026 this paragraph said no pricing page had been found and that third-party figures were the only ones available.
muse-spark-1.3, labelled “Not used to improve our products.”: $1.25 input, $0.15 cached input and $4.25 output per million tokens, with a 1M-token context window.Meta, Meta Model API, read at source 22 Sep 2026: “muse-spark-1.3 Not used to improve our products. 1M $1.25 $0.15 $4.25”.muse-spark-1.3-contributor, labelled “Used to improve our products.”: $0.10 input, $0.002 cached input and $0.20 output per million tokens, with the same 1M-token context window.
Meta, Meta Model API, read at source 22 Sep 2026: the table rows read “muse-spark-1.3 Not used to improve our products. 1M $1.25 $0.15 $4.25” and “muse-spark-1.3-contributor Used to improve our products. 1M $0.10 $0.002 $0.20” (context window, then input, cached input and output per million tokens). The same table is on Meta’s Muse Spark 1.3 page, read the same day. A snapshot: re-query, do not trust it.
The cheap tier is paid for partly with your data. The contributor rate is less than a tenth of the full rate on input and on output, and the only difference Meta states is the label. The page read does not say exactly what is used or how, so assume anything you send on that tier may be used to improve Meta’s products.
Meta is not yet in the four-vendor price table, which was built before this price appeared. The rule on this site is that a price comes from the company charging it.
What Muse Spark is for
Meta positions 1.3 around long agent runs rather than chat. Its own framing: the model “tracks context and prior results, works through messy or conflicting inputs, and asks for input when needed”, and it is “Trained for long-horizon, agentic workflows”.
On multimodal input: “Muse Spark perceives video, images and documents, and its visual reasoning runs through a real execution environment instead of scripted steps.”
The same model family powers Meta’s consumer agent, which is covered on what Meta Muse is, and the open-weight sibling you can run yourself is on how to run Muse Glimmer locally.
What this page could not verify
- When the token price appeared. It was not found on 16 Sep 2026 and was on Meta’s pages by 22 Sep 2026; the date in between is not known.First-hand: until 22 Sep 2026 this item read “Muse Spark’s token price. Not published on Meta’s developer site.”
- Any quality or speed claim. All are Meta’s own, none tested here.
- Whether $0.01 covers every image size. The page gives one figure without qualifying it.Meta, AI developer site, read in a browser 16 Sep 2026: “Muse Image is now on Meta Model API at $0.01/image”, stated without a size qualifier.
- Rate limits or free tiers. Not stated on the page read.
Meta, Products & Solutions for AI Developers, read in a browser on 16 September 2026 — the site is script-rendered and returns no text to an automated request, so it was read in a browser pane and the wording transcribed.
Meta, Meta Model API and Muse Spark 1.3, and the Muse Code blog post of 31 August 2026, all read at source on 22 September 2026.
Comparison prices from Google and xAI, both read at source the same day as the first check.