Google announced Gemini 4 Argon on September 30 as a model for long, multi-step work in software engineering, enterprise research and cybersecurity defense. Its rollout is initially limited to trusted cyber defenders in the Fairwind Program; Google says developers, enterprises and consumers will receive access later.
The central technical change is a maximum output of 1 million tokens, up from 64,000. Google says that headroom lets the model generate hundreds of thousands of tokens in a single trajectory while working through complex problems. This is an output limit, not a stated context-window size; the announcement does not specify the latter.
Google says its teams are already using Argon for debugging, algorithm design and large code migrations. Projects range from smaller libraries to more than 800,000 lines in the Fuchsia Zircon kernel. For the libgav1 video decoder, agents replaced 32,000 lines of SIMD code in an existing Rust port; Google reports that the result ran 2.7 times faster than that port while producing identical video output. The company says critical rewrites undergo automated and manual audits, emulation testing and review before production deployment.
In Google’s published evaluations, Argon scored 77.9% on the long-horizon software-engineering benchmark DeepSWE v1.1, 51.3% on AutomationBench and 91.7% on the long-video test LVBench. Google also claims leading results on the Vals Index and specialized finance and legal evaluations. These are vendor-reported benchmark results, not independent confirmation of performance in a particular workplace.
Cybersecurity is a major focus: Google says Argon can autonomously find, validate and patch critical software vulnerabilities. It tied for first with 68% on CWE-bench v1, according to the announcement. Trusted defenders and Google’s internal teams are due to receive a version without cyber guardrails, while broader release preparations include defenses against misuse and indirect prompt injection, monitoring for actions that exceed user intent, and hardened sandbox environments.
Practical context: For prospective customers, availability remains the immediate constraint. Google is collecting feedback from early testers and participating in the US government’s voluntary pre-release access process. It has not announced a date for broad availability, so Argon should not yet be treated as a generally accessible API or consumer Gemini model. Google says paid API customers and Google AI Ultra subscribers will be first as access expands.
When the commercial launch begins, introductory pricing is set at $2 per million input tokens and $10 per million output tokens, with cached input priced 95% below the standard input rate. After the introductory period, those rates are scheduled to rise to $4 and $20 respectively. Because Google gave no precise launch date, these prices describe a future service rather than one broadly available today.
| Period | Input per 1M tokens | Output per 1M tokens | Cached input |
|---|---|---|---|
| Introductory | $2 | $10 | 95% off the input price |
| After the introductory period | $4 | $20 | Not specified |
Sources
Event date: 2026-09-30. Primary source date: 2026-09-30.
