Google Announced Gemini 4, but You Can’t Use It Yet
Matthew Leo · Published October 2, 2026 · AI
Google has announced Gemini 4 Argon, its new flagship AI model, but most developers and consumers cannot use it yet. Access is initially limited to selected cybersecurity organizations in Google's Fairwind program, and the company has not provided a date for a public release.
That gap matters because the announcement includes prices, benchmark scores and broad claims about coding, legal and financial work. Those details describe Google's planned product, not a service Canadians can sign up for today.
What Google is releasing now
Google says Argon is built for long-running work across software engineering, enterprise research and cyber defence. Its September 30 announcement gives the model a one-million-token context limit and says it can coordinate multi-step tasks for extended periods.
The first outside users are “trusted cyber defenders” participating through Fairwind. Google says their feedback will help it improve safeguards before access expands to paid API customers, enterprises and Google AI Ultra subscribers.
The company is also participating in a voluntary U.S. government process for pre-release model access. That is separate from the new White House AI safety accord, which relies largely on company commitments rather than an enforcement agency.
No Canadian launch date appears in the announcement. Google published a French-Canadian version of the post, but localized information is not proof that the model is available in Canada.
The pricing is public before the product is
Google lists an introductory API price of US$2 per million input tokens and US$10 per million output tokens. After the introductory period, it says those prices will rise to US$4 and US$20 respectively. Cached input is listed at a 95 per cent discount from the input rate.
There are still missing details. Google has not said when billing begins, how long the introductory period lasts, whether the same rates will apply in every region or what access limits customers will face. Canadian businesses will also pay in Canadian dollars through their billing arrangement, so the actual cost will vary with exchange rates and taxes.
The published rate is useful for planning, but it should not be treated as a current Canadian price until the API is actually offered here.
Google’s benchmarks need outside testing
Google reports strong results on coding, cybersecurity and professional-work benchmarks. It also describes internal projects in which Argon helped optimize a quantum-computing routine, identify memory savings across data centres and assist with large C++-to-Rust migrations.
Those examples come from Google. They are not independent tests, and several involved human review, automated testing or controlled internal systems. Reuters noted that Argon led competing models on some of Google's chosen measures but trailed on two of the four coding benchmarks included in the release.
The fairest conclusion is that Google has presented a capable-looking model with incomplete outside evidence. Independent evaluators need access before buyers can compare reliability, latency, tool use and real project costs.
Why cybersecurity access comes first
Google says Argon can find and patch vulnerabilities, which makes security teams useful early testers. The same ability creates risk if a model can autonomously explore systems or write exploit code.
The company says it is strengthening sandbox isolation, monitoring agent behaviour and testing resistance to prompt injection before a wider rollout. That caution is reasonable. It also means the restricted launch is part of the product story, not a small availability footnote.
Organizations selected for early access should still use narrow permissions, separate test environments and human approval for changes. A model that can operate for a long time and call tools can make a small instruction error much more consequential.
What Canadian users can do now
Consumers should not change Gemini subscriptions to get Argon; Google has not said that an existing plan provides access. Developers should not build production schedules around the announced prices until the API appears in their account and region.
Canadian companies evaluating the model can prepare a fixed test set now: real tasks with confidential details removed, known correct answers and cost limits. They should record what competing models achieve on the same work so Argon can be compared rather than judged from Google's demonstrations.
The next meaningful update will be actual availability. Until Google names regions, dates and access conditions, Gemini 4 Argon is an announced model in a controlled rollout, not a generally available Canadian product.
Tags: Gemini 4, Google, AI models, Canada