Google Gemini 4 Argon: New Model for Complex Professional Tasks
Google DeepMind announces Gemini 4 Argon, a frontier-class model built for complex, long-running professional tasks with 1M token context window.
👁 117 reads
Update, October 13, 2026: Gemini 4 Argon is now generally available: in the paid tier of the Gemini API (no free API tier) and the Gemini platform for business, in Google Antigravity for AI Ultra subscribers, and in the Gemini app and AI Mode in Search for Google AI Pro and Ultra subscribers (rolling out gradually). The API price of 2 USD input / 10 USD output per 1M tokens is introductory and runs until November 30, 2026; from December 1 it is 4 / 20 USD. Current prices, availability and our test
Google DeepMind announced Gemini 4 Argon on September 30, 2026, a new frontier-class system built for complex, long-running professional tasks. For now, only select cybersecurity specialists have access through the Fairwind program; Google states it will make the model available to developers, businesses, and individual users as quickly as the safety verification process allows.
Argon offers a context window of 1 million tokens, which Google describes as industry-leading. Starting pricing is $2 per million input tokens and $10 per million output tokens; cached tokens are 95% cheaper than the standard input price. Google is also participating in the voluntary government pre-release process in the United States.
Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →What Argon Is Already Doing Inside Google
The model is already being actively used by thousands of Google employees. The most tangible results are visible in three areas. Agents based on Argon analyzed telemetry from server fleet profiling and autonomously deployed memory optimizations that freed over 300 TiB after the first deployment - target savings are estimated at 500 TiB to 1 PiB. In quantum algorithm optimization, Argon improved a published benchmark by 40% within minutes, helping researchers reduce space-time resources (qubits × gates) of application bottlenecks.
In parallel, agents are conducting migrations of C/C++ codebases to Rust - from tens of thousands of lines in libraries like re2 and libgav1 up to over 800,000 lines of the Zircon kernel in the Fuchsia operating system. Due to the criticality of these systems, every change undergoes rigorous automated and manual auditing, emulation testing, and review before production deployment.
The decision to gradually phase in access to the model - rather than an immediate public launch - signals that Google is learning from criticism accompanying earlier Gemini releases and this time is prioritizing test credibility over speed to capture headlines.
More on this from QUANTSLANT
benchmarks·Gemini 4 Argon vs GPT-6 Astra vs Claude Opus 5.5 BenchmarksGemini 4 Argon wins 13 of 19 benchmarks against GPT-6 Astra and Claude Opus 5.5, with launch pricing at one-fifth of Astra's cost. Full comparison with API costs.