AI Pricing This Week: Gemini 4 Argon, Muse Wonder, Oct 12-18

Short answer. Between October 12 and 18, 2026, Google and Meta each released a model at $4.00 / $20.00 per million tokens. Google sells Gemini 4 Argon at $2.00 / $10.00 until November 30. This site corrected three of its own readings. Gemini Omni Flash Preview shuts down on October 22, and six OpenAI models on October 23.

In plain words. Two new models arrived, and two shutdowns are days away. Prices are per million tokens. A token is a short piece of text, about three-quarters of a word. If you only pay for a chat app, the part for you is the new AI subscriptions section.

This is the first weekly summary on this site. It lists what this site recorded between Monday, October 12 and Sunday, October 18, 2026: new models, price changes, our own corrections, shutdowns coming up and the pages added. Every fact comes from this site’s dataset and change log, each with the provider page it was read on and the date; the comments between them are ours.

The week in three lines. Google and Meta each released a model at the same $4.00 / $20.00 list price, and Google sells its own at half that until November 30. This site corrected three of its own readings, one of which had shown a promotion as a price cut for a week. And two shutdowns are days away: Gemini Omni Flash Preview on October 22 and six OpenAI models on October 23.

New models

Meta’s model page reads “muse-wonder-1.0 Available in Model API and Muse Code. 1M $4.00 $0.20 $20.00”, the figures being context, input, cached input and output; Meta’s documentation adds “Muse Wonder is in preview.” Google’s pricing page prints Argon’s input as “$2.00 through November 30, 2026. $4.00 starting December 1, 2026.” Argon entered this site one day after its release and Muse Wonder two days after. Both are compared, with a month of the same workload on each, in Gemini 4 Argon vs Muse Wonder.

One open-weights model was added: Muse Spark Open, which Meta’s model card describes as “an 816B total/42B active parameter open weights model with a context window of 1M tokens.” Two hosts price it, and both print the same figure on October 17: $1.25 input and $4.25 output per million tokens on Together AI and on Fireworks AI. Only one of them calls its price temporary. Fireworks’ model page, read again on October 19, says: “Limited preview: Available on serverless as a limited preview through October 29, 2026, on US-only capacity , with the usual 50% US-only premium waived.” Together’s pricing page, read the same day, prints “Muse Spark Open | $1.25 $0.15 (cached) | $4.25 |” with no promotion beside it.

Price changes, and our own corrections

No provider changed a list price this week in this site’s data. All three entries in the price-change log for these days are this site’s own, and each one says so in its first line.

Kimi K3 on Together AI, October 12: a promotion read as a cut. From October 6 to October 12 this site showed Kimi K3 on Together AI at $2.70 input and $13.50 output and logged it as a 10% price cut. It was a promotion. On October 12 Together’s pricing page labelled the row “PROMO” with “Promotion valid until October 11, 2026” and printed the list price beside it, $3.00 and $15.00. This site publishes list prices, so the row went back to $3.00 / $15.00 and the entry of October 6 was replaced. The Kimi K3 page shows the corrected figures. This is the mistake of the week, and it was ours: for six days this site showed a promotional price as if it were the list price, and called it a cut.

Gemini 3 Flash Preview, October 13: an output price we did not have. The row had an input price ($0.50) and a cache price ($0.05) but no output price, because it had been read on Google Cloud’s Vertex AI page, whose standard table prints only input rows for this model. Google’s Gemini API pricing page, the source of every other Gemini row, prints an output price of $3.00 per million tokens. The Gemini 3 Flash Preview page now cites that page. This site does not know since when Google has printed the figure.

Muse Spark 1.1, October 14: a price we had missed. Since September 16 this site had shown Muse Spark 1.1 as having no published price, because its model page has no pricing table. Meta’s “Pricing and rate limits” page, which we had not read, lists it at $1.25 input, $0.15 cached and $4.25 output, with the line “Muse Spark 1.1 is Standard mode only.” The Muse Spark 1.1 page now carries that price.

A new section: AI subscriptions

On October 17 this site opened AI subscriptions: seven products (ChatGPT, Claude, Google Gemini, Microsoft Copilot, Perplexity, Grok and Mistral), each plan quoted from the maker’s page with its region, currency, tax and the day it was read. ChatGPT, Perplexity and Grok refuse scripts and were read by hand from Spain; Claude has a US dollar reading and a euro reading from Spain, and Google Gemini a US dollar reading and a reading from Spain with its personal plans in euros and its Workspace plans in US dollars, listed separately and never converted. Three comparisons were published the same day: ChatGPT Plus vs Claude Pro, Google AI Pro vs ChatGPT Plus and Perplexity Pro vs ChatGPT Plus.

Four other posts went out during the week: what a month of the same open model costs on each host (October 12), the Claude Sonnet 5.5 cache-read cut and every Gemini Flash generation’s price (October 13), and the Argon and Muse Wonder comparison (October 14).

What comes next

Dates as the providers print them on their deprecation and pricing pages, all listed on Deprecations and the price-change log:

• October 22: Gemini Omni Flash Preview shuts down. Google’s line: “gemini-omni-flash-preview | June 30, 2026 | October 22, 2026 | gemini-omni-1.1-flash |”, the last field being its replacement.
• October 23: OpenAI shuts down o1, o1-pro, o3-mini, GPT-4, GPT-4 Turbo and GPT-3.5 Turbo, each with a named replacement, for example “October 23, 2026 | o1-2024-12-17 | o1 | gpt-5.6-sol |”. The cost of moving is worked out in OpenAI’s October 23 shutdown.
• November 21: OpenAI says GPT-5.6 Sol’s “promotional pricing is available at least through November 21, 2026” and publishes no price for after it.
• November 30: the last day of Argon’s introductory price; from December 1 it is $4.00 input and $20.00 output. The same day, Anthropic retires Claude Sonnet 4.5: “November 30, 2026 | claude-sonnet-4-5-20250929 | claude-sonnet-5-5 |”; see the cost of moving off Sonnet 4.5.
• December 11: OpenAI shuts down GPT-5, GPT-5 mini and o3. Its deprecations page lists “Dec 11, 2026 | gpt-5-2025-08-07 | gpt-5.6-sol |”, “Dec 11, 2026 | gpt-5-mini-2025-08-07 | gpt-5.6-terra |” and “Dec 11, 2026 | o3-2025-04-16 | gpt-5.6-sol |”, the last field being the replacement.
• December 31: the promotional price of Gemini 3.8 Flash and Gemini 3.6 Flash ends. On January 1, 2027 input goes from $0.75 to $1.50 and output from $3.75 to $7.50, the list prices Google prints next to the promotional ones.

Update, October 23, 2026: Google’s deprecations page, read again on October 23, 2026, still printed “gemini-omni-flash-preview | June 30, 2026 | October 22, 2026 | gemini-omni-1.1-flash |”, and Google’s pricing page, read the same day, no longer listed Gemini Omni Flash Preview. Google’s Gemini Omni Flash model page, also read on October 23, 2026, printed “Model versions: Stable: gemini-omni-1.1-flash Shut down: gemini-omni-flash-preview”.

Update, October 26, 2026: Fireworks now prints “Available on serverless on US-only capacity at standard serverless pricing, with no US-only premium.” and “Muse Spark Open runs on US-only capacity. The 50% US-only premium is waived, so the rates in that row are what you pay.” Same price.

What we will watch

The Kimi K3 error came from reading one figure without the label beside it. The lesson we take is to read the label next to a price as well as the price, and if we miss one again it will be logged here the same way. On December 1 we will check that Google’s page has moved Argon to the list price it announced, rather than assume it. Fireworks’ limited preview of Muse Spark Open runs through October 29; after that date we will report the price its page prints, and we do not assume what it will be. For GPT-5.6 Sol, OpenAI’s wording is “at least through November 21”, so the date may pass with no change; we will report what the page says, not what we expect it to say. The subscription pages are re-read by hand every Tuesday, and any change to a plan will be recorded with the date it was seen. We do not know why any of these prices are what they are, and this summary does not guess.

How this was put together

The facts above were listed by this site’s weekly script from its own data files: the models dataset, the price-change log, the deprecation records, the subscription data and the posts published in the week. Every price and date was read on the provider’s own page, on the date given on each model or log page. Model prices are list prices per million tokens in US dollars, except where a promotional price is named as such, as with Gemini 4 Argon through November 30.

Frequently asked questions

Which new AI models did this site add between October 12 and 18, 2026?

Gemini 4 Argon, released by Google on October 13 at $2.00 input and $10.00 output per million tokens through November 30, and Muse Wonder, released by Meta on October 12 at $4.00 and $20.00. Both were added on October 14, and the open model Muse Spark Open got its page on October 17.

What did this site correct about its own prices that week?

Three readings: Kimi K3 on Together AI was a promotion published as a cut and is back at its $3.00 / $15.00 list price, Gemini 3 Flash Preview gained its missing $3.00 output price, and Muse Spark 1.1 got the $1.25 / $4.25 price Meta lists on its pricing page. None of the three was a price change by the provider.

Which AI models shut down in the next few weeks?

Gemini Omni Flash Preview on October 22, 2026, and OpenAI's o1, o1-pro, o3-mini, GPT-4, GPT-4 Turbo and GPT-3.5 Turbo on October 23, 2026. Claude Sonnet 4.5 follows on November 30, 2026, and OpenAI's GPT-5, GPT-5 mini and o3 on December 11, 2026.