Google launches Gemini 4 Argon after DeepMind shakeup and months of delays
Google finally dropped its flagship Gemini 4 Argon model for deep work, but the cheap API pricing vanishes after November 30.
Models · Source: Reuters
What happened
Google just released Gemini 4 Argon. It is their new flagship AI model built for complex, multi-step tasks. The launch comes after months of delays and a massive leadership shakeup at the DeepMind lab. CEO Demis Hassabis stepped aside, and several leaders of the Gemini model left the company. Google even completely scrapped its planned Gemini 3.5 Pro model, which was originally supposed to launch in June. Now, they are trying to regain ground against rivals like OpenAI and Anthropic.
Argon is available right now for developers via the API in AI Studio. Consumers paying for Google AI Pro and Ultra can access it through the Gemini app and AI Mode in Search. Google claims the model ranks first on the Vals Index for economic work like finance, coding, and law. Early tests look strong. Google used it internally to untangle six million lines of legacy browser feature code in Chrome. They finished an architectural refactor across hundreds of files in weeks instead of months.
The company is also opening its Fairwind Program to hundreds of new partners. This gives vetted cybersecurity defenders a special version of Argon with the cyber guardrails ripped out. In September, Google said it was only providing this to select partners, but now it is expanding. Meanwhile, Google is handing ten thousand scientists six months of free access to its Antigravity coding tool to push research forward. Early enterprise customers testing the model include Deloitte, LexisNexis, Ambience Healthcare, and Isomorphic Labs.
Key facts
- $2 — Starting price per million input tokens before November 30
- 35% — Increase in agentic code submissions using Argon in Antigravity
- 10,000 — Scientists getting six months of free access to Antigravity
- 95% — Discount applied to cached input tokens
Why it matters
Builders finally get access to a Google model that can handle serious agentic workflows. Argon is not just a chatbot. It is built for deep work like massive codebase refactoring and quantum chemistry optimization. Google claims teams using Argon in its Antigravity tool increased agentic code submissions by over 35 percent while cutting rollbacks by nearly a third. That means the model writes code that actually works in production. Researchers even used it to make the open-source xTB quantum chemistry package up to 2.3 times faster at predicting molecular structures.
The second-order effect is a brutal pricing war with a ticking clock. Google is charging an introductory rate of $2 per million input tokens and $10 per million output tokens. They are also offering a massive 95 percent discount on cached input tokens. But that base rate doubles after November 30. Google is buying market share right now to steal developers back from competitors. If you are building high-volume agentic applications, you have a very short window to test if Argon is worth the permanent price tag.
For builders
Lock in cheap API rates before November 30
Argon launches at $2 per million input tokens and $10 per million output tokens. Google will double these rates after November 30. Builders need to run their benchmarks now before the massive discount disappears forever.
Leverage unchained models for cybersecurity tools
Google is expanding its Fairwind Program to hundreds of partners. Vetted security startups can access a version of Argon without cyber guardrails. This is a massive advantage for founders building offensive security and penetration testing products who usually fight model refusals.
Automate massive codebase refactoring and optimization
Argon excels at untangling legacy code. Google used it to separate six million lines of Chrome code and speed up Gemma inference by nearly three times. Engineering teams can use Argon to pay down technical debt that was previously too expensive or risky to touch.
My take
I think Google desperately needed this win after scrapping Gemini 3.5 and losing top DeepMind talent. Argon looks incredibly powerful for agentic coding, but doubling the API price in six weeks is a hostile move. They are forcing us founders to rush our evaluations instead of earning our trust.
Original reporting: Reuters. This is my rewrite and opinion.