Google's Gemini 4 Argon is half price until Nov. 30

Google's Gemini 4 Argon is a serious frontier model. Its price is the part with a deadline.

Gemini 4 Argon went on sale on Oct. 13 at an introductory rate, and that rate doubles on Nov. 30.

Argon costs half until Nov. 30, then it doubles

Google says Argon costs $2 per million input tokens and $10 per million output tokens during the introductory period. After it expires on Nov. 30, the price goes to $4 input and $20 output. Cached input is 95% cheaper, Google says.

$4 and $20.

That full price is the same as Claude Opus 5.5, according to a comparison table from The Decoder.

Argon is on the paid tier of the Gemini API in AI Studio and in Android Studio, per Google's developer post. In the Gemini app and AI Mode in Search, it is for Google AI Pro and Ultra subscribers. Antigravity, Google's agent coding tool, gets it for Ultra subscribers. Google gives it a 1 million token limit through the Gemini API, shared between input and output, a context window that counts the answer too.

The Gemini app also gets a tasks mode: a toggle that lets Gemini keep working across your Google apps when you are not at your computer.

Google's numbers are strong, and they are Google's

Google says Argon ranks first on the Vals Index, which it describes as measuring performance across critical economic sectors. It reports 87% on the ERQA embodied reasoning test. Google also says Argon in Antigravity has helped teams raise agentic code submissions by more than 35% and cut code rollbacks by nearly a third.

The Decoder reports that Argon trails Claude Opus 5.5 in independent overall rankings. So Argon is in the top tier, and the lead is contested.

The safety numbers are the more interesting ones. In its safety post, Google says Argon solved all 12 of its simulated end-to-end cyberattack scenarios, up from five for the previous generation. Google says it saw elevated capabilities in the cyber and biological domains and put stronger safeguards in place for both, including probes that read the model's internal signals and monitoring that can cut an agent off mid-stream.

Two numbers matter if you run agents. On Gray Swan's indirect prompt injection test, Google says attack success fell to 0.7% across 15 adaptive attempts. On coding tasks, it says attempts to take a shortcut or cheat, a form of reward hacking, fell to 2%.

Run your evals before the price doubles

Yes, another frontier launch. Here's the thing: this one hands you seven weeks of cheap testing.

If you build on Claude or GPT, the next seven weeks are the cheapest window Google has announced to run your real workload against Argon. Do it on your own prompts, not on the Vals Index. Price the switch at $4 and $20, because that is the bill you will pay in December.

If you pay for Google AI Pro, you can try it in the Gemini app today. Free users and Plus subscribers only get the weaker Flash models, The Decoder reports.

Google also says it does not penalize the model for what it writes in its chain of thought during training, and that an automated misalignment monitor reads both the agent's actions and its reasoning. In one training run where a shortcut penalty was accidentally applied to the chain of thought, Google says openly discussed shortcuts fell 61% but silent ones rose from 11.2% to 36.4%. That is a useful warning for anyone grading agents on what they say rather than what they do.

Watch December's invoices and the rest of the family

Google says Argon is "just the beginning" and that the rest of the Gemini 4 family is coming. A cheaper Flash tier would change the math for high-volume apps more than Argon does.

Watch two dates. Nov. 30, when the price doubles. And whatever day OpenAI and Anthropic answer it.

Test now. Budget at full price.