Google has begun rolling out Gemini 4 Argon, its new frontier model, first to trusted cyber defenders through the Fairwind Program rather than a full public launch. In a Sept. 30, 2026 Google DeepMind blog post, SVP Koray Kavukcuoglu said Argon targets complex software engineering, enterprise knowledge work, and cybersecurity defense while Google strengthens safeguards ahead of broader access.
Phased Gemini 4 Argon access and pricing
Google said Argon is entering a phased release while the company participates in the U.S. government’s voluntary pre-release model access process. Introductory API pricing is listed at $2 per million input tokens and $10 per million output tokens, with cached input tokens discounted 95%, before later rising to $4 and $20 respectively. Paid API customers and Google AI Ultra subscribers are slated to get access after early testing, according to the company.
Engadget’s Oct. 1 coverage noted Google positioning Argon against other frontier systems and highlighted Artificial Analysis scores that put Argon’s Intelligence Index in line with OpenAI’s GPT-6 Astra at a lower reported cost per task under discounted pricing.
Long-horizon coding and enterprise workflows
Google says Gemini 4 Argon expands output length to 1 million tokens, up from 64K on prior generations, to support longer multi-step tasks. Internally, Google reported Argon agents helping free more than 300 TiB of data-center memory and assisting large C/C++ to Rust migrations, including work tied to the Fuchsia Zircon kernel. On DeepSWE v1.1, Google cited a 77.9% score for long-horizon software engineering; Argon also ranked first on Zapier’s AutomationBench at 51.3%, per the announcement.
Readers following Google’s enterprise AI push may also want our earlier look at Gemini Enterprise for financial services and last week’s coverage of OpenAI’s Dots agents at DevDay 2026.
Cyber defense first, then wider release
Google trained Argon heavily for defensive cybersecurity and is releasing a version without cyber guardrails to trusted Fairwind partners so they can find, validate, and patch vulnerabilities. Google said cloud security firm Wiz used Argon in its Scan for Good program and identified a critical exposure in healthcare software missed by earlier frontier models. On CWE-bench v1, Google said Argon tied for first at 68%.
Before a broad consumer and developer rollout, Google said it is hardening misuse defenses, prompt-injection resistance, misalignment monitoring, and sandboxed evaluation environments. That caution sits alongside intensifying AI product scrutiny, including our report on the FTC’s OpenAI and Anthropic probe.