Ing. Calogero Bono
Google has unveiled Gemini 4 Argon, a model designed for complex tasks in coding, finance, law, cybersecurity and science. It is available right away via API, in the Gemini app and in Antigravity, depending on your subscription. The introductory price is $2 per million input tokens and $10 per million output tokens. It matters because Google is pitching it for long, multi-step work rather than quick chat, and it claims measured results on real projects.
A few terms, before we get into the facts. Token is the unit models use to read and write text, and it is the basis of pricing. Antigravity is Google's environment for working with AI agents. Refactoring means reorganizing code without changing its behavior. Lean 4 is a tool that checks mathematical proofs line by line. Zero-day is a security flaw still unknown to those who should fix it.
What is Gemini 4 Argon and where can it be used?
According to the official Google blog, Argon is available globally to developers and businesses via API in AI Studio and on the Gemini platform. Google AI Ultra subscribers find it in Antigravity, while Ultra and Pro users have it in the Gemini app and in AI Mode in Search.
Google claims it outperforms other frontier models in tests and that it ranks first on the Vals Index, which measures performance in areas such as finance, code, law and tax. This is a claim made by the company, not verified by us.
Other models are also in the same family, such as Gemini 4 Hydrogen, designed for speed and agents.
How much does Gemini 4 Argon cost?
Google defines the launch price as introductory. It costs $2 per million input tokens and $10 per million output tokens. Input tokens stored in cache cost 95% less.
The introductory term suggests the price list could change, but the source does not say when. If you are designing a service, it is wise to calculate costs with a margin.
If you want to understand how to integrate these models into your products, the Gemini API guide is a good starting point.
What results does Google claim on code?
Google reports internal results. In Antigravity, Argon reportedly increased agents' code proposals by over 35% and cut rollbacks by almost a third.
Examples cited
- Chrome. A refactoring across hundreds of files, with months of work completed in weeks, on a legacy folder of about 6 million lines.
- Agent Development Kit. Optimizations found overnight, with 40% less time spent preparing long sessions and cold imports 21% faster.
- Gemma. A 2.85x inference speedup on A100 with vLLM, from about 37,900 to over 108,000 tokens per second, with no loss of quality.
These are cases chosen and described by the company itself. Those working with coding agents can compare them with tools such as Cursor.
What can Argon do in security and science?
On the security front, Google announces the extension to hundreds of other partners of the Fairwind program. Wiz uses it for the free Scan for Good initiative. According to the source, it helped close a flaw that risked exposing government data and a file download bypass that gave administrator access to a healthcare facility.
In science, a Google Research post describes Argon and the /teamwork multi-agent system on open problems in mathematics and theoretical computer science, with proofs verified in Lean 4. Among them, the Courtade-Kumar conjecture from 2013. The work is carried out alongside human experts.
A related theme is formal code verification, as in AlphaProof Code.
Limits to know
Almost all the data comes from Google and its partners, with no independent verification. The tests are internal or run on private data, such as those from Gamow Labs. The official source text we have is truncated, so some results are missing. There is also no date set for the end of the introductory price, and we do not know how Argon behaves on your own workloads.
What changes for those working in digital
For developers, Argon is a model to try on long refactoring, code review and optimization, where the work spans several steps. For companies, the discounted input price with cache makes workflows with repeated contexts, such as legal or financial documents, attractive. Those who manage infrastructure can evaluate its use in security audits, always with human review. You need to measure costs and quality on a real case before choosing. Infrastructure matters too, as shown by the launch of the new TPUs.
Our take
Argon stands out for its method more than its score. Google shows concrete cases, with numbers and public repositories, which partly allows checking. The limit remains internal data, chosen by the company selling the product. The real test will be that of external teams. Moreover, the introductory price is a commercial lever and should be read that way. The comparison with competitors on always-on agents, such as Perplexity Teammates, will tell us a lot.
Frequently asked questions
How much does Gemini 4 Argon cost?
It costs $2 per million input tokens and $10 per million output tokens, at an introductory price. Cached input tokens cost 95% less.
Where can I use Gemini 4 Argon?
You can use it via API in AI Studio and on the Gemini platform. Ultra subscribers get it in Antigravity, Ultra and Pro in the Gemini app, and in AI Mode in Search.
Is Argon the best on the market?
Google says so, ranking it first on the Vals Index. We have no independent verification in the sources, so the figure should be taken as the company's claim.
What is Antigravity's /teamwork?
It is Google's multi-agent system for long problems. It splits the work among specialized agents and uses verification tools, such as Lean 4 or software tests, to check the solutions.
Is it useful for cybersecurity?
According to Google, yes, and Wiz used it to find and fix flaws. It is, however, a tool for experts. Every report must be verified by people.
What to watch after Argon's launch price
In the coming months three things matter – the final price list, independent evaluations and the arrival of the /teamwork tools. These points will show whether Google's results hold up outside its labs.
- Watch for announcements on the end of the introductory price and redo your cost calculations.
- Look for third-party comparisons on coding tests and real-world workloads.
- Follow updates on /teamwork in Antigravity to know when it will be available to everyone.
