Skip to content

DirectoryTools

Gemini 4 Argon logo

Gemini 4 Argon

Google's frontier Gemini model for long coding, legal, finance and science work, on the paid API.

Visit Gemini 4 Argon

Agents & ChatPaidWe checked the prices on ai.google.dev on 3 Nov 2026.

Gemini 4 Argon homepage screenshot

Recaply review

Gemini 4 Argon review: strong knowledge work at Opus 5.5's price

Gemini 4 Argon is Google DeepMind's frontier model for long coding, legal, finance and science work, with one 1M-token window for input and output, sold on the paid Gemini API and Google AI plans.

$4
Per million input tokens at Google's standard price; $20 per million out
1M
Tokens shared by input, thinking and output, with long decode up to 1M out
77.9%
On DeepSWE v1.1 long-horizon coding in Google's tests, against 74.2% for Claude Opus 5.5
19.6%
On Harvey's Legal Agent Benchmark, where no other model in Google's table passed 6.7%

What is Gemini 4 Argon?

Gemini 4 Argon is Google DeepMind's flagship reasoning model, built to keep working through long, multi-step jobs in software engineering, finance, legal work, science and cyber defense. Input, thinking and output share one 1M-token window, and with long decode a single response can run to 1M tokens instead of stopping at 64K. It reads text, images, video, audio and PDFs.

Google has released it generally: developers pay per token on the Gemini API's paid tier, and Google AI Pro and Ultra subscribers use it in the Gemini app. Google has since announced Gemini 4 Carbon, which builds on Argon and scores higher in Google's tests, but only vetted cyber defenders can use Carbon so far.

Where can I use Gemini 4 Argon?

On the Gemini API's paid tier, in the Gemini app with Google AI Pro or Ultra, and without cyber guardrails for vetted defenders in Google's Fairwind Program.

Developers
The Gemini API paid tier in Google AI Studio and Android Studio, at $4 / $20 per million tokens.
Subscribers
The Gemini app and AI Mode in Search on Google AI Pro and Ultra; Ultra adds Google Antigravity and AI Studio.
Enterprises
Gemini for business, the API through Google Cloud, and Google Antigravity.
Cyber defenders
Fairwind Program partners get a version without cyber guardrails, after background checks and with phishing-resistant MFA.

Who is Gemini 4 Argon best for?

Gemini 4 Argon is best for engineers, analysts and researchers who need a model to work through long coding, legal, finance or science tasks over very large inputs.

  • Long coding and migration jobs

    Google reports 77.9% on DeepSWE v1.1, and its Chrome team used Argon to untangle core page handling from a 6-million-line folder of browser features, finishing months of work in weeks.

  • Legal and finance research

    It scored 19.6% on Harvey's Legal Agent Benchmark and 65.4% on Vals Finance Agent v2, the top marks in Google's table.

  • Very long documents and video

    With a 1M-token window, it scored 84.2% on GraphWalks prompts of 256K to 1M tokens and 91.7% on LVBench long-video understanding, ahead of every rival Google listed.

  • Cyber defense teams

    Fairwind Program partners get it without cyber guardrails to find, validate and patch vulnerabilities, and Google is adding hundreds more partners.

What's included with Gemini 4 Argon?

A 1M-token window with long decode, three thinking levels, text, image, video, audio and PDF input, agent tools including computer use, prompt injection resistance and cheap cached input.

1M shared window and long decode
Input, thinking and output share 1,048,576 tokens, and long decode runs one response past 64K, up to 1M tokens, in the background.
Thinking levels
Low, medium and high settings trade reasoning depth for speed and cost; medium is the default.
Multimodal input
It reads text, images, video, audio and PDFs, with the top scores on Chartography and LVBench in Google's table.
Agent tools
Function calling, computer use, code execution, file search, Google Search and Maps grounding, URL context and structured outputs are all supported.
Prompt injection resistance
Google reports the lowest attack success rate in Gray Swan's indirect prompt injection test, 0.7% after 15 attempts.
Cheap cached input
Repeated input is cached automatically and billed at $0.20 per million tokens, 95% below the standard input price; explicit caching is not offered.

How much does Gemini 4 Argon cost?

Gemini 4 Argon costs $4 per million input tokens and $20 per million output tokens on the Gemini API's paid tier, and it is included in Google AI Pro at $19.99 a month and Google AI Ultra from $99.99 a month.

Gemini API, paid tier$4 / $20Per million input / output tokens
  • Cached input $0.20 per million tokens
  • Not on the Gemini API free tier
  • Batch, Flex and Priority processing
Google AI Pro$19.99Per month
  • Argon in the Gemini app and AI Mode in Search
  • 4 times the Free plan's usage limits
  • 5 TB of cloud storage
Google AI Ultra$99.99Per month
  • Highest Argon access in the Gemini app
  • Argon in Google Antigravity and AI Studio
  • 5 times Pro's usage limits; 20 times for $199.99

Prices are in US dollars. On the Gemini API every thinking and output token is billed, and Google Search grounding includes 5,000 free requests a month across Gemini 3 and newer models, then $14 per 1,000. Data sent on the paid API tier is not used to improve Google's products. Vetted cyber defenders can apply to Google's Fairwind Program for a version without cyber guardrails. Full pricing on ai.google.dev. We checked these prices on .

How do readers rate Gemini 4 Argon?

No one has rated Gemini 4 Argon yet. Be the first.

No ratings yetThe score and its stars show once 3 readers have rated Gemini 4 Argon.
Your rating

Sign in to rate. One rating per account.

Your rating of Gemini 4 Argon

Gemini 4 Argon pros and cons

It leads Google's table on knowledge work and long context at Opus 5.5's price, but it trails Opus 5.5 and GPT-6 Astra on some coding tests and has no free API tier.

Pros

  • Leads Google's table on knowledge work: 68.9% on the Vals Index, 65.4% on Vals Finance Agent v2 and 19.6% on Harvey's legal benchmark, ahead of GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5.
  • Hacker News users who have used it since release call it thorough and strong at math, writing and sticking to a goal, and a few now prefer it to Claude Opus 5.5.
  • Same standard price as Claude Opus 5.5, $4 / $20 per million tokens, with cached input billed 95% below the input price.
  • A 1M-token window shared by input and output, and long decode that lets one response run to 1M tokens, 16 times the earlier 64K limit.

Cons

  • Trails rivals on some coding tests in Google's own table, 57.4% on Terminal-Bench 4.0 against 66.4% for Opus 5.5, and several Hacker News users still find Opus 5.5 stronger for coding.
  • Google AI Pro subscribers on Hacker News call Argon's chat usage limits in the Gemini app low, and Antigravity access is aimed at Ultra subscribers.
  • Not on the Gemini API free tier, so testing it through the API means paying per token from the start.
  • Google has already announced Gemini 4 Carbon at the same standard price, which beats Argon on almost every test in Google's table, though only cyber defenders can use it so far.

What are the best Gemini 4 Argon alternatives?

Claude Opus 5.5 for coding at the same price, GPT-6 Astra for FrontierSWE and computer use, and Gemini 3.8 Flash for a cheaper Google model with a free tier.

Questions about Gemini 4 Argon

What is Gemini 4 Argon's context window?

Gemini 4 Argon has a single window of 1,048,576 tokens that input, thinking and output share. Replies stop at 65,536 tokens by default, and long decode with background execution lets one response run up to 1M tokens minus your input.

Who is Gemini 4 Argon for?

Gemini 4 Argon is best for engineers, analysts and researchers who need a model to work through long coding, legal, finance or science tasks over very large inputs.

Is Gemini 4 Argon on the Gemini API free tier?

No. Google's pricing page lists Gemini 4 Argon on the paid tier only, at $4 per million input tokens and $20 per million output tokens, with thinking tokens billed as output. Gemini 3.8 Flash is the closest Google model with a free tier.

How much does Gemini 4 Argon cost?

Gemini 4 Argon's paid plans are Gemini API, paid tier at $4 / $20 per million input / output tokens, Google AI Pro at $19.99 per month and Google AI Ultra at $99.99 per month.

What are the best Gemini 4 Argon alternatives?

The closest alternatives on Recaply are Claude Opus 5.5, GPT-6 Astra and Gemini 3.8 Flash.

Where can I try Gemini 4 Argon?

On its official website, blog.google.

Can I use Gemini 4 Argon in the Gemini app?

Yes, with a paid plan. Google AI Pro at $19.99 a month and Google AI Ultra from $99.99 a month include Gemini 4 Argon in the Gemini app and AI Mode in Search, with Ultra getting the highest limits. The free Gemini plan and Google AI Plus do not list it.

What is the difference between Gemini 4 Argon and Gemini 4 Carbon?

Gemini 4 Carbon builds on Argon, has the same standard price of $4 / $20 per million tokens and scores higher on almost every test in Google's table. Argon is on general sale, while Carbon is open only to vetted cyber defenders in Google's Fairwind Program so far.

How do I apply for the Fairwind Program?

Fill in the interest form on DeepMind's Fairwind Program page. Google prioritizes governments, critical infrastructure operators and core technology platforms, runs background checks, and requires phishing-resistant MFA with access limited to security teams.

Our verdict

Gemini 4 Argon is a real frontier option for knowledge work and long-context jobs, and you can now buy it.

Google's own table puts it first on legal, finance, long-context and video tests at Opus 5.5's price, and early users praise its thoroughness, but it trails Opus 5.5 and GPT-6 Astra on terminal and FrontierSWE coding.

Skip it for
Teams whose main job is agentic coding, and anyone who needs a free API tier.
Check first
Test it against your current model on your own tasks at each thinking level, since every score so far is Google's.