Gemini 4 Argon’s $2/$10 Price Is Half the Later Rate

Google says Gemini 4 Argon will launch at $2 per million input tokens and $10 per million output tokens. A footnote on Google’s own announcement of 30 September 2026 says $4 and $20 apply after an introductory period it does not define.

The Price Has a Footnote

Google’s announcement of Gemini 4 Argon says it “will launch at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off input token price.” The word introductory carries a footnote.

The footnote reads: “After the introductory period expires, the price of $4 per 1M input tokens and $20 per 1M output tokens will apply.” The post does not say how long the introductory period lasts.

This publication’s own arithmetic: the launch price is half the post-introductory price on both input and output. If the 95% discount applies to the price in force, cached input is $0.10 per million tokens at launch and $0.20 after.

What Was Posted on X

Google DeepMind posted on 16 October that four Gemini 4 Argon agents beat Minecraft from scratch in 27 hours, working in a sandbox through code and text only. It says they wrote 2,783 programs, split up jobs over chat, shared notes and debugged errors together without human help.

Maxence Faldor’s post, linked by DeepMind, describes the setup: four Gemini 4 Argon agents in one shared Minecraft world with the goal of defeating the Ender Dragon, which fell after 27 hours and 2,783 self-written programs. It links to his longer X article on the run.

Perplexity CEO Aravind Srinivas wrote on 16 October that Gemini 4 Argon set a new state of the art on the Wide-And-Deep-Research (WANDR) benchmark and is available to all Perplexity users in fast search and in Computer (agent) mode.

Perplexity’s own account, two minutes earlier, said Gemini 4 Argon set a new record on WANDR and is available in Perplexity for Pro and Max subscribers. The two posts describe availability differently: the company account names Pro and Max, while Srinivas says all users.

The How To Prompt account, on 16 October, retold DeepMind’s Minecraft run for a general audience: four Gemini 4 Argon agents given one objective, defeating the ender dragon, controlling avatars through an open-source library, talking in game chat, building shared memory and writing code. It is a summary of the demonstration, not a new source.

Google’s Logan Kilpatrick, on 14 October, posted a conversation with Sergey Brin, Josh Woodward, Tulsee Doshi and Melvin Johnson about Gemini 4 Argon and how Google got there.

An account’s 16 October post transcribes Brin from that conversation: “even before Gemini 4, back with Flash 3.6, 3.7, 3.8 — we started getting onto that RSI treadmill where the researchers and the model work hand-in-hand on the next version”, adding that internally Google is “already using a couple of models ahead”. The quotes are the poster’s transcription.

Google’s Logan Kilpatrick posted on 13 October that Gemini 4 Argon was rolling out globally in the Gemini API, to Pro and Ultra subscribers in the Gemini app, and to Ultra subscribers in Antigravity. He wrote that the model had been tested by over 100,000 people at Google over the previous month.

Google’s Antigravity account said on 15 October (US time) that Gemini 4 Argon was rolling out in Antigravity to paid Google AI Pro subscribers, with a separate weekly-only Argon quota for Pro on top of the existing Gemini bucket for 3.8 Flash. It said demand had been high and capacity was scaled deliberately.

Who Can Use It Now

Google says Argon is rolling out to a set of trusted cyber defenders through its Fairwind Program, and that it is engaged in the U.S. government’s voluntary process for pre-release model access while it gradually expands access.

It says broad availability will start with paid API customers and Google AI Ultra subscribers, and gives no date beyond “as soon as possible.” The $2 and $10 price is therefore one Google has announced for a launch that has not yet happened.

The Benchmarks Google Chose

Google says Argon sets a new state of the art on DeepSWE v1.1 at 77.9%, ranks first on Zapier’s AutomationBench at 51.3%, and is state of the art on LVBench at 91.7%. On CWE-bench v1, which evaluates fixing security vulnerabilities, it says Argon ties for first place with 68%.

It also says Argon is the leading model on the Vals Index, and shows similarly leading performance on Vals Finance Agent v2 and Harvey’s Legal Agent Benchmark, without giving scores for those three.

All of these are Google’s reported results. This publication did not open any of the leaderboards and did not repeat any test.

Google’s Own Workloads

Google says Argon agents analyzing fleet-wide profiling telemetry freed up over 300 TiB of memory once rolled out, with an estimated 500 TiB to 1 PiB in total savings, and that in one quantum-computing example the model beat a published baseline by 40% in minutes.

It says Argon agents are migrating C and C++ code to Rust, from tens of thousands of lines in core libraries up to 800,000 lines or more for the Fuchsia Zircon kernel, and that for the libgav1 video decoder the result runs 2.7 times faster than the existing Rust port with identical video output.

Google adds that these rewrites are undergoing automated and manual auditing, emulation testing and review before they reach production.

One Million Output Tokens

Google says it is expanding Argon’s output token limit to an industry-leading 1 million tokens, up from the previous 64,000. This publication’s own arithmetic: that is about 15.6 times the earlier limit.

The post does not say what that costs. This publication’s own arithmetic: one maximum-length response of 1 million output tokens comes to $10 at the launch price and $20 at the later price, ignoring input and cached tokens.

The Structural Read

The footnote reads: “After the introductory period expires, the price of $4 per 1M input tokens and $20 per 1M output tokens will apply.” The post does not say how long the introductory period lasts, and it does not give the date the later prices take effect.

The launch has not happened yet. Google says Argon is rolling out to a set of trusted cyber defenders through its Fairwind Program, and that broad availability will start with paid API customers and Google AI Ultra subscribers, with no date beyond “as soon as possible.”

The output limit changes what one response can cost. Google says it is expanding Argon’s output token limit to 1 million tokens, up from 64,000. The article’s own arithmetic puts one maximum-length response at $10 at the launch price and $20 at the later price, ignoring input and cached tokens.

Three Implications

TWO PRICES, ONE ANNOUNCEMENT Google says the launch price is $2 and $10 per million tokens, and its footnote says $4 and $20 apply after the introductory period. This publication’s own arithmetic is that the launch price is half the later price on both input and output.

THE DATES ARE MISSING The length of the introductory period, the date of broad availability and the effective date of the $4 and $20 prices are not given.

EVERYTHING HERE IS GOOGLE’S CLAIM The prices, the benchmark scores and the internal results are Google’s own claims. This publication verified none of them and did not open any third-party leaderboard.

What Is Not Established

The length of the introductory period and the date of broad availability are not given, and neither is the effective date of the $4 and $20 prices. The benchmark scores and internal results are Google’s claims and were not independently verified.

For the earlier piece on Argon’s cyber guardrails, see this publication’s coverage of the Fairwind Program. Nothing here is investment advice.

Business Engineer Framework

The Map of AI — The Stack Beneath the Applications

The Map of AI places more than 200 companies across nine layers of the stack, from silicon to application. It shows the layers that sit beneath the applications people use every day.