Google announced Gemini 4 Argon on 30 September 2026, but most developers cannot use it yet. Outside Google, the model is available only to vetted cybersecurity teams. Here is who has access, how it performs and what it will cost.
Also Read: Write-Ahead Log (WAL): How Databases Survive Crashes - How To's
Who can use Gemini 4 Argon today
Access runs through Google's Fairwind Program, SiliconANGLE reports. Fairwind opened on 3 September and has signed up more than 650 organisations, including CrowdStrike and Palo Alto Networks.
Paying developers and Google AI Ultra subscribers are next in line. Google has not given a date.
Why Google is holding it back
Google says a model at this level needs a phased release. It is taking part in the US government's voluntary pre-release review and is still hardening safeguards against misuse.
Fairwind members get a version with the cyber guardrails removed. That lets defenders find and fix vulnerabilities, which is the same capability an attacker would want.
Also Read: Jobs: How Much More
How it performs
These are Google's own published benchmarks, as reported by SiliconANGLE.
| Benchmark | Gemini 4 Argon | Claude Opus 5.5 | GPT-6 Astra |
|---|---|---|---|
| DeepSWE v1.1 (long software tasks) | 77.9% | 74.2% | 74.1% |
| AutomationBench (business workflows) | 51.3% | 42.5% | |
| CWE-bench v1 (vulnerability fixes) | 68% | 68% |
Argon led 12 of the 18 benchmarks in Google's charts. Opus 5.5 still leads Terminal-Bench 4.0.
Also Read: News and Nvidia $6 trillion market cap
The output limit is 1 million tokens. Earlier Gemini models stopped at 64,000.
What Google has built with it
- Moving C and C++ code to Rust, including the 800,000-line Zircon kernel in Fuchsia.
- Rewriting 32,000 lines of the libgav1 video decoder as safe Rust. The result runs 2.7 times faster than the earlier port.
- Finding wasted memory across its data centres. The sweep has freed more than 300 tebibytes so far.
Pricing
| Input, per million tokens | Output, per million tokens | |
|---|---|---|
| Launch pricing | $2 | $10 |
| After launch pricing ends | $4 | $20 |
Cached input is priced 95% lower. The later rates match what Anthropic charges for Opus 5.5.
Also Read: OpenAI Apologises After AI Agents Breach Australian Sites - News
The regulatory backdrop
On the same day, the FTC confirmed an industry-wide probe into AI labs including OpenAI and Anthropic, according to reporting that cites Reuters. Security review is becoming part of the release process for frontier models.
What this means for developers
- Do not plan a launch around Argon until Google names a date.
- Budget on the $4 and $20 rates, not the launch prices.
- The 1 million token output limit suits long migrations and refactors, so line up candidates now.
For more on AI agents writing production code, read our Filament Blueprint review and What Is Jev?. OpenAI's side of the race is in our DevDay 2026 developer guide.
Also Read: ASOS Hack: What Happened and What Customers Should Do
FAQ
Can I use Gemini 4 Argon now?
Only if your organisation is in Google's Fairwind Program. Developers and AI Ultra subscribers are next, with no date set.
How much does Gemini 4 Argon cost?
$2 per million input tokens and $10 per million output tokens at launch, rising to $4 and $20 later.
