Gemini 4 Argon raises output limit to 1M tokens

Google has announced Gemini 4 Argon, a frontier model it describes as built to sustain deep reasoning across long, multi-step workflows. The company announced it on 30 September 2026 in a post on blog.google, signed by Koray Kavukcuoglu, senior vice president at Google DeepMind. It is not generally available: the model is rolling out to a set of trusted cyber defenders through Google's Fairwind Program, with wider access to follow.

A 1M output token ceiling, up from 64K

The change Google leads with is the output token limit. Argon can generate up to 1 million tokens in a single trajectory, against 64K previously, which Google calls industry-leading. Output tokens are what the model writes back, as opposed to the input it reads, so the ceiling governs how much reasoning and text one run can produce before it stops. Google's stated reason is headroom: with room to generate hundreds of thousands of tokens in one pass, the model can work through a hard problem without being cut off.

Google reports frontier performance in software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defense. The numbers it cites are its own or from benchmark owners: 77.9% on DeepSWE v1.1 for long-horizon software engineering, first place on the Vals Index, which weights finance, coding, legal and tax work by contribution to U.S. GDP, and 51.3% on AutomationBench, Zapier's benchmark for end-to-end execution across business functions. On LVBench, for long video understanding, Google reports 91.7%.

Cyber defenders get Argon without cyber guardrails

For trusted defenders and Google's internal teams, Argon is being released without cyber guardrails, so they can use its full cybersecurity defense capabilities. Google says the model can autonomously find, validate and patch critical software vulnerabilities, and reports a top score of 68% on CWE-bench v1, tied for first. Wiz, the cloud security company, is using Argon through its Scan for Good initiative; Google says the model uncovered a critical vulnerability in healthcare software used by hospitals worldwide that earlier frontier models had missed.

Everyone else waits. Google says it is iterating on guardrails and gathering feedback from early testers before making Argon available to developers, enterprises and consumers, starting with paid API customers and Google AI Ultra subscribers. It also says it is taking part in the U.S. government's voluntary process for pre-release model access. The introductory price is $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 after that period, with cached input at 95% off the input price.

The concrete fact to hold: a 1M output token limit, and a first cohort of cyber defenders rather than enterprise customers.