Google announces Gemini 4 Argon, rolling out first to cyber defenders
Google's new frontier model targets long, complex work in coding, law, finance and cybersecurity. It goes to trusted cyber defenders first, with developers and paid API customers next.

Google has announced Gemini 4 Argon, which it calls its new frontier model. The announcement, published on 30 September 2026 by Koray Kavukcuoglu of Google DeepMind, positions Argon for long, multi-step work in three areas: real-world software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defence.
Who can use it, and when
Argon is not generally available yet. Google is rolling it out in phases:
- First: a group of trusted cyber defenders through its Fairwind Program. Google says these defenders, and its own internal teams, get Argon without cyber guardrails.
- Next: developers, enterprises and consumers, starting with paid API customers and Google AI Ultra subscribers. Google gives no date beyond "as soon as possible".
Pricing
| Introductory price | After the introductory period | |
|---|---|---|
| Input | US$2 per 1M tokens | US$4 per 1M tokens |
| Output | US$10 per 1M tokens | US$20 per 1M tokens |
Cached input tokens are priced at 95% off the input price.
What's new
- A much bigger output limit: 1 million output tokens, up from 64,000, so the model can reason and write at great length in one go.
- Coding: Google reports a state-of-the-art 77.9% on DeepSWE v1.1, a benchmark of long software-engineering tasks. Internally, Argon agents are migrating C and C++ codebases to Rust. For Google's open-source libgav1 video decoder, Google says the result runs 2.7 times faster than the earlier Rust port.
- Knowledge work and video: Google says Argon leads the Vals Index for finance, coding, legal and tax work, and scores 91.7% on the LVBench long-video benchmark.
- Security: Google says Argon can find, validate and patch software vulnerabilities on its own, and ties for first place on CWE-bench v1 with 68%.
All of these figures are Google's own.
Safety before wider release
Google says it is strengthening four sets of safeguards before broad release: refusing help with cyber and chemical, biological, radiological and nuclear attacks; resisting prompt injection; monitoring the model's reasoning for misalignment; and hardening the sandboxes used for testing.
What it means for builders here
For most developers in Malaysia, nothing changes today: Argon isn't open to general API users yet. If you already pay for the Gemini API, it is worth watching for access, and noting that prices are in US dollars and double after the introductory period. Benchmarks are a starting point, not a guarantee. Test it on your own tasks before moving production work.
The full announcement is on Google's blog.


