Google's Gemini 4 Argon Goes to Cyber Defenders First, at $2 Per Million Input Tokens

Google DeepMind announced Gemini 4 Argon on Sept. 30 and said it will reach a set of trusted cyber defenders first, through what it calls the Fairwind Program. Developers, enterprises and consumers follow "as soon as possible," according to Google's announcement, which was written by Koray Kavukcuoglu, SVP of Google DeepMind and Chief AI Architect.

As of Thursday morning the post gives no date for that wider release. It says paid API customers and Google AI Ultra subscribers come first once Google finishes iterating on guardrails.

What Argon costs

Google set an introductory price of $2 per million input tokens and $10 per million output tokens. Cached input tokens carry a 95% discount on the input price. The post lists the regular price after the introductory period as $4 and $20.

Tier	Input, per 1M tokens	Output, per 1M tokens
Introductory	$2	$10
Regular	$4	$20

Google did not say how long the introductory period lasts. Argon's output limit rises to 1 million tokens, up from 64,000 on the previous model, the post says. It does not state an input context window.

Artificial Analysis, an independent benchmarking site, lists a 1 million token context window, text and image input, and text-only output.

What the independent index shows

Artificial Analysis scored Argon (High) 53 on its Intelligence Index, ranking it 8th of 223 models. Running the index cost $1.99 per task, which ranks 77th of 223. Argon generated 110 million tokens during the evaluation, ranking 74th for verbosity.

The benchmark numbers in Google's post are vendor-supplied. Google reports 77.9% on DeepSWE v1.1, 68% on CWE-bench v1 for vulnerability remediation, 51.3% on AutomationBench and 91.7% on LVBench. All four are Google's own figures, and the post names no outside party that ran them.

One commenter on Hacker News called the results "big number" but wrote that benchmarks feel saturated and that they would wait for hands-on use.

Fairwind and the cyber guardrails

Google said it will give trusted defenders and internal teams a version of Argon without cyber guardrails. The post says Argon "can autonomously find, validate, and patch critical software vulnerabilities."

The post says little about who qualifies for Fairwind. It does say Argon agents are migrating C and C++ code to Rust inside Google, from tens of thousands of lines in libraries such as re2 and libgav1 up to more than 800,000 lines in the Fuchsia Zircon kernel.

Google also said misalignment mitigations monitor Argon's chain of thought and actions and stop execution when necessary. The company said it took precautions against feeding monitoring findings back into training.

What is still open

Rival pricing is the obvious yardstick. OpenAI priced GPT-6.1 Sol at one fifth of Astra's rates, and the same week it detailed a distillation campaign aimed at extracting protected model reasoning. Google's post says Argon's reasoning is monitored, not that it is hidden from users.

Google has not said when Argon leaves the trusted-defender tier. It has not said how many organisations are in Fairwind or what the introductory price becomes if the rollout slips. The post says Google will keep collecting feedback from early testers while it iterates on guardrails.