3-Line TL;DR

  • Google opened Gemini 4 Argon to everyone on October 13: the API, the Gemini app and AI Mode for Pro and Ultra subscribers, and Antigravity for Ultra only
  • In my Gemini app use it made things up noticeably less and reasoned better than the previous version, but each question took about 5% of my AI Pro usage meter
  • It is clearly frontier-level now and a successor called Carbon is reportedly in testing, but on a $19.99 plan I can't use it enough to rely on it

Out at last, two weeks after the announcement

  • Google announced Argon on September 30 for a group of cyber defenders only
  • My last post complained that there was still no date for anyone else
  • On October 13 Google made it generally available
  • Developers get it through the Gemini API at $2 input / $10 output per 1M tokens until November 30, then $4 / $20
  • Pro and Ultra subscribers get it in the Gemini app and AI Mode in Search

Fewer made-up answers, and the numbers agree

AA index chart (Opus 5.5 58, Argon 53, Astra 53, Gemini 3.1 Pro Preview 30) and hallucination rates (Argon 15%, Astra 51%, Sol 54%)

Source: Artificial Analysis, Argon at high effort, checked October 14, 2026

  • In my Gemini app use, Argon made things up noticeably less than the previous version, and its reasoning felt stronger
  • Artificial Analysis measured a 15% hallucination rate, the lowest among models scoring 45 or more on its index, against 51% for GPT-6 Astra
  • That rate counts how often a model answers wrongly instead of admitting it doesn't know
  • Part of the gain is saying "I don't know" more: Argon got 50% of the questions right, Astra 63%
  • On the overall index Argon scores 53, tied with Astra and up from 30 for Gemini 3.1 Pro Preview
  • Claude Opus 5.5 still leads at 58

5% per question is about 20 questions

Twenty blocks showing that about 5% of the AI Pro usage meter per question lasts roughly 20 questions

HSL chart — my own usage meter reading; limit rules from Gemini Help, checked October 14, 2026

  • On my AI Pro plan, each Argon question took about 5% off the usage meter
  • At that rate a full meter lasts about 20 questions
  • Google counts usage by compute, so prompt complexity, model, features and chat length all change the cost
  • Pro limits refresh every five hours, with a weekly cap on top
  • So 20 is my rough math for questions like mine, not a fixed Google number
  • Twenty questions is enough to be impressed, not enough to work with it

Antigravity says Ultra only

Table of Argon access by plan: Pro gets the Gemini app and AI Mode but not Antigravity, Ultra gets both, plus API pricing

Source: Google announcement, Google One US plans; USD before tax, checked October 14, 2026

  • In Antigravity, Google's coding agent app, Argon is listed for AI Ultra subscribers only
  • Pro users there see it greyed out as "currently available to Ultra plan subscribers" (Hacker News)
  • Ultra starts at $99.99 a month in the US, five times the $19.99 Pro plan
  • Long coding sessions are where usage burns fastest, and that is exactly where Pro is shut out
  • The other route is the API, paying per token

Frontier status saves face, and Carbon is next

Timeline: Argon announced Sep 30, Carbon testing reported Oct 9, Argon general release Oct 13, Carbon release date unknown

HSL diagram — Google announcements; Carbon from a Business Insider report via secondary coverage

  • With Argon, Google is back near the top of the index after Gemini 3.1 Pro Preview's 30
  • Business Insider reported on October 9 that Google staff are testing another Gemini 4 build, codenamed Carbon, on an internal coding tool
  • One employee said it "feels like Opus 5.5" for coding, which is one opinion with no benchmark
  • Google has not said whether Carbon becomes an Argon update or a separate model
  • A frontier model out the door and a stronger build already in testing is enough to save face
What I want What I would do now
Try Argon in chat Pro at $19.99 works; budget about 20 questions per window
Argon in Antigravity for coding Ultra from $99.99, or the API
Better value Watch for Carbon and the API price change after November 30

Community Reactions

  • A Hacker News user said Argon isn't available to Pro in Antigravity, while Opus 5.5 and Sol or Astra work on $20 plans, and most people won't pay $100 a month just to try it (Hacker News)
  • Another called it "pretty insane" that Pro users don't get even a taste in Antigravity, noting the app includes a sample of Opus 5.5 (Hacker News)
  • One commenter called Argon a new leader on cost versus intelligence and said Google is back in the race (Hacker News)
  • Another described subscription fatigue now that OpenAI, Anthropic, Meta and Google all have frontier models (Hacker News)

Q&A (Field Notes)

  • Q. Is this the long-awaited Gemini 4 Pro?
    • Google calls it Gemini 4 Argon, and its announcement does not use a "4 Pro" name
  • Q. Do free or AI Plus users get Argon?
    • The launch post names only Pro and Ultra for the Gemini app, so check your own account before paying for anything
  • Q. Is the API cheaper than a subscription for coding?
    • It depends on volume; Artificial Analysis put one index task at $1.99 at the introductory price and $3.98 at the standard price
  • Q. Should I upgrade to Ultra just for Antigravity?
    • Only if Argon-powered coding is your main job and $80 more a month is fine; otherwise I would wait for Carbon or a price change