Google DeepMind launched Gemini 4 Argon on September 30, 2026, positioning the new frontier model directly against systems from OpenAI and Anthropic with a 1-million-token output limit and a significantly lower hallucination rate. Initially distributed through the Fairwind Program to select cybersecurity defenders and United States government partners, the model ties for top general intelligence on independent benchmarks while triggering a more than 3% rise in Alphabet shares in after-hours trading.
Benchmark Performance and Architecture Updates
Gemini 4 Argon matches the GPT-6 Astra at the pinnacle of general intelligence on the Artificial Analysis Intelligence Index, edging out the GPT-6.1 Sol by one point according to platform data. Google DeepMind structured the system around deep reasoning and multi-step problem solving for long-form workflows. The model expands the output token ceiling from 64 mil to 1 million, allowing single-run analysis of extended video files and complex software engineering projects.
Independent evaluation places the model’s hallucination rate at 15%, contrasting with 51% for GPT-6 Astra and 54% for GPT-6.1 Sol. While the system demonstrates strong capabilities in corporate economic impact, agent automation, and cybersecurity defense, it trails Anthropic and OpenAI models in Terminal utilization, specialized coding agents, and scientific code generation.
https://x.com/sundarpichai/status/2105387952478277979
Controlled Rollout via the Fairwind Program
Sundar Pichai confirmed the release through a post on X, while Google DeepMind released accompanying technical documentation detailing safety protocols. Access remains restricted during the initial phase to participants in the Fairwind Program, which includes designated cybersecurity defenders and US government partners.
Did you know?
Alphabet shares rose more than 3% in after-hours trading following the announcement of Gemini 4 Argon, reflecting positive market reaction to Google’s reentry into top-tier benchmark competition after months of competitive pressure from OpenAI and Anthropic throughout 2026.
Pichai emphasized that the phased distribution prioritizes responsible development, with wider availability for developers and enterprise clients dependent on feedback gathered from the initial security reviews. Safety measures built into the architecture include defenses against prompt injection and safeguards to prevent misuse in cyber or biological threat vectors.
https://x.com/GoogleDeepMind/status/2105388084154056939
Token Pricing Structure
During the introductory launch window, Google set the cost for Gemini 4 Argon at $2 per million input tokens and $10 per million output tokens, alongside a 95% discount for cached input tokens. Following the initial promotional period, standard rates will adjust to $4 per million input tokens and $20 per million output tokens.
Access limits and benchmark performance of Gemini 4 Argon
Who has access to Gemini 4 Argon right now?
The model is currently available exclusively to select cybersecurity defenders and United States government partners participating in Google’s Fairwind Program.
What is the maximum token output for Gemini 4 Argon?
The system supports an output limit of 1 million tokens, up from the 64 mil token limit found in previous generations.
How does Gemini 4 Argon perform on hallucination benchmarks?
According to the Artificial Analysis Intelligence Index, Gemini 4 Argon records a 15% hallucination rate, compared to 51% for GPT-6 Astra and 54% for GPT-6.1 Sol.
Related reading