โ๏ธ Blog โบ ๐ค Tech & AI โบ AI models
Gemini 4 Argon, Muse Wonder and Maia 300: the AI week, explained
By The Trends Incoming team ยท Published 2026-10-14
Checked Nov 20, 2026. Google's Gemini API pricing page still lists Gemini 4 Argon at $2 per million input tokens and $10 per million output tokens through November 30, then $4 and $20 starting December 1. Argon is still not on the free tier. The October launch story is below, with the price-sheet comparison that page did not have.
Some weeks in AI are quiet. The week of October 12 was not. In about 48 hours, Meta launched a closed frontier model, Google shipped Gemini 4 Argon, and Microsoft posted the Maia 300 chip plan. Here is what each company actually published, re-read on November 20.
Google Gemini 4 Argon
Google released Gemini 4 Argon on October 13. The launch post says it is available to developers through the Gemini API, to enterprises, and to Google AI Pro and Ultra subscribers in the Gemini app and AI Mode in Search. The headline spec is a 1 million-token combined input-output limit. Introductory API pricing in that post is $2 per million input tokens and $10 per million output tokens until November 30, then $4 and $20.
The Verge reported that Google claims Argon beats rival models on coding, cybersecurity, science, finance and law benchmarks, and that it cuts "hedging, repetition, and the cliches." Benchmark claims from the company that made the model still need a second opinion. We have not added an independent score.
On the pricing page checked November 20, Argon is one of four Gemini 4 names, and the token rates are not all the same:
| Model on the Gemini API price sheet | Input / output per 1M tokens | Free tier |
|---|---|---|
| Gemini 4 Carbon | $2 / $10 through Nov 30, then $4 / $20 | No |
| Gemini 4 Argon | $2 / $10 through Nov 30, then $4 / $20 | No |
| Gemini 4 Neon | $3 / $15, no intro window listed | No |
| Gemini 4 Hydrogen | $1.50 / $7.50 | Yes, for input and output |
Google calls Carbon its most capable reasoning model and says Argon "delivers Gemini 4 intelligence with faster responses and lower token spend per turn" than Carbon on coding, agent, math and science tasks. Same sticker price, Google's claim of fewer tokens per turn. Cached input on both Carbon and Argon is listed at $0.10 per million through November 30 and $0.20 after, and the page says explicit caching is not supported. The pricing page does not date when Carbon, Neon or Hydrogen were added, so this table is a November 20 reading of the sheet, not a launch calendar.
Keep reading
AI models hubThe frontier race, and the money question hanging over it. Image policy
Open-source modelsThe open-weight side, which Muse Wonder stepped off. Image policy
Tech and AIThe rest of the category, from chips to outages. Image policy
OpenAI's 2026 incidentsWhat OpenAI has published about agents and safeguards. Image policy
Meta Muse Wonder
A day earlier, Meta announced Muse Wonder and Muse Wonder Full, built for coding, reasoning and agents. Unlike Meta's earlier open-weight Llama models, neither is available to download or self-host. Meta's own Model API pricing page, checked November 20, still lists muse-wonder-1.0 at $4 per million input tokens and $20 per million output tokens, with cached input at $0.20. Contributor mode, the cheaper tier Meta offers on Muse Spark, is marked not available for Muse Wonder. The pricing page does not list a Muse Wonder Full API rate. The October launch post still limits Full to Muse Code subscribers.
Microsoft Maia 300
Microsoft's October 12 blog, re-read November 20, says Maia 300 will begin production deployments in 2027 and that Microsoft expects it to deliver 40% more tokens per megawatt than leading GPU-based systems. In a 96-accelerator configuration, the same post says it delivers up to 7 times the rack-scale token throughput of Maia 200. An earlier version of this page, relying on a newsletter summary, called that a "40% better performance per megawatt." The blog's own phrase is tokens per megawatt. Maia 200, the generation already out, is described as serving production workloads including Microsoft Copilot. Maia 300 is the one that has not shipped.
Why it's trending
Because the leaderboard just got crowded. The Verge framed Argon as Google's first major frontier release since late 2025, after months of trailing rivals, and called it "a reminder of how quickly the balance of power can shift in AI." Meta going closed is a big philosophical shift too. And chips matter because whoever controls the compute controls the cost.
The backstory
The frontier AI race has mostly been framed as a contest between a few labs. That October week showed more players sprinting at once. Google was trying to win back the lead. Meta has rebuilt its AI effort under Meta Superintelligence Labs, which TechRepublic says took about 14 months. Microsoft is building its own silicon so it isn't entirely dependent on one supplier.
What's next
- December 1, 2026. Google's price sheet moves Argon and Carbon from $2 and $10 to $4 and $20 per million tokens. Neon and Hydrogen do not show that intro window.
- Argon is still off the free tier. Hydrogen is the Gemini 4 name that is not.
- Muse Wonder Full still has no API price on Meta's pricing page. Maia 300 is still a 2027 production date, not a chip you can rent.
Where to follow it
The Gemini API pricing page, Meta's Model API pricing page and Microsoft's Maia blog are the primary pages. The Verge and TechRepublic covered the October launch week. Hashtag links below are conversation, not the price sheet.
Related next reads
AI models hubWhere this week sits against the rest of the model race. Image policy
Open-weight modelsWhat you can still download, after Muse Wonder closed. Image policy
OpenAI security incidentsThe 2026 reports OpenAI published about its own agents. Image policy
Tech and AIThe category page for everything else in this beat. Image policy
Quick answers
What is Gemini 4 Argon?
Google's frontier model released October 13, 2026, with a 1 million-token limit. On November 20, 2026 the Gemini API price sheet still listed it at $2 per million input tokens and $10 per million output tokens through November 30, then $4 and $20.
Is Meta's Muse Wonder open source?
No. Unlike Llama, neither Muse Wonder model can be downloaded or self-hosted. Meta's pricing page, checked November 20, 2026, still lists muse-wonder-1.0 at $4 per million input tokens and $20 per million output tokens.
What is Maia 300?
Microsoft's third-generation in-house AI chip. Microsoft's October 12, 2026 blog says production deployments begin in 2027 and that it expects 40% more tokens per megawatt than leading GPU-based systems.
Is Gemini 4 Argon on the free tier?
No. Google's pricing page says Argon is not available on the free tier. Gemini 4 Hydrogen is the Gemini 4 name on that page with free input and output.
Follow it
The trend card
Google Gemini 4 Argon
Google's new frontier model, out Oct 13 with a 1M-token window. It's pitched as less hedgy and less clichรฉ.
Sources
- Google blog: Gemini 4 Argon
- Gemini API pricing
- Meta Model API pricing
- Microsoft blog: Maia 300
- The Verge
- TechRepublic: Muse Wonder
By The Trends Incoming team ยท Fact-checked 2026-10-31. Dates change, so double-check with the official source.