Google has announced Gemini 4 Argon, its latest flagship AI model, skipping the delayed Gemini 3.5 Pro. The model features an industry-leading 1 million output token limit and is designed for complex enterprise workflows. Google is executing a phased rollout, limiting initial access to trusted cybersecurity partners in its Fairwind Program and U.S. government agencies. While Google claims benchmark leads in legal and financial tasks, independent tests show a close race with OpenAI and Anthropic.
Release and distribution of Gemini 4 Argon
- ▪Google is initially rolling out Gemini 4 Argon to trusted cybersecurity defenders through its Fairwind Program, a cybersecurity initiative, and internal teams without cyber guardrails to allow full access to its defensive capabilities
- ▪Google is participating in the U.S. government's voluntary pre-release model access process, introduced under the Trump administration, to evaluate Gemini 4 Argon before a wider public release
- ▪Google skipped the release of its previously promised Gemini 3.5 Pro model, deciding instead to focus on Gemini 4 Argon after development delays
- ▪Google announced Gemini 4 Argon on September 30, 2026, as its latest flagship artificial intelligence model designed for complex enterprise workflows
Internal use cases and capabilities
- ▪Google employees used Gemini 4 Argon to migrate C/C++ codebases to Rust, including over 800,000 lines in the Fuchsia OS Zircon kernel
- ▪Google employees used Gemini 4 Argon agents to identify data center memory optimizations, freeing over 300 TiB of memory
Expanded output capabilities
- ▪Gemini 4 Argon supports an output limit of 1 million tokens per response, a 16-fold increase from the 64,000-token limit of previous Gemini models such as Gemini 3
- ▪Google is adding a Long Decode Continuation feature to the Gemini API to prevent long responses from hitting timeouts
Benchmark performance and comparisons
- ▪Gemini 4 Argon leads outright on 12 of 18 benchmarks disclosed by Google, including the Harvey Legal Agent Benchmark and AutomationBench
- ▪Gemini 4 Argon scored 68.9% on the Vals Index, ranking first ahead of OpenAI's GPT-6 Astra and Anthropic's Claude Opus 5.5
- ▪Independent testing by Artificial Analysis tied Gemini 4 Argon with OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1 on the Artificial Analysis Intelligence Index
- ▪Gemini 4 Argon trails OpenAI's GPT-6 Astra on the FrontierSWE v2 benchmark, scoring 55.0% compared to Astra's 65.5%
API pricing and discounts
- ▪Google announced introductory API pricing for Gemini 4 Argon at $2 per million input tokens and $10 per million output tokens
- ▪Google said following Gemini 4 Argon's introductory $2-per-million-input/$10-per-million-output pricing period, its pricing will rise to $4 per million input tokens and $20 per million output tokens
Debatable claims
- ▪Google is justified in releasing Gemini 4 Argon to cybersecurity partners without guardrails
- ▪Google's phased rollout of Gemini 4 Argon is an appropriate way to balance safety and innovation
- ▪The U.S. government should rely on voluntary self-regulation agreements with AI developers
Story comments
Loading comments…