TL;DR: Google announced Gemini 4 Argon on Sept. 30, 2026, as its new top-tier AI model, but limited initial access to trusted cyber defenders through its Fairwind Program while it completes added safety testing.

Google introduces Gemini 4 generation

Alphabet's Google announced a new top-tier artificial intelligence model to anchor its Gemini 4 generation of models on Wednesday, Sept. 30, 2026, according to Reuters. The company called the model "Argon."

Reuters reported from San Francisco that the announcement was a fresh bid to catch up to rivals Anthropic and OpenAI in the AI race. A company spokesperson told Reuters that Argon is larger in size than Google's previous line of advanced "Pro" models.

In its own announcement, Google described Gemini 4 Argon as its new frontier model built to sustain deep reasoning across complex, long-horizon workflows. According to the official Google blog post by SVP Koray Kavukcuoglu, the model delivers frontier performance in complex workflows across real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense.

Kavukcuoglu is identified by Google as SVP, Google DeepMind and Chief AI Architect, Google. According to The Verge, Kavukcuoglu became the new boss of DeepMind in August.

Why access is limited at launch

Google is not making Argon broadly available at first. According to the Google blog, Argon is rolling out to a set of trusted cyber defenders through its Fairwind Program.

The company said safely releasing frontier capabilities at this level requires a phased approach. According to Google, it is actively engaged in the U.S. government's voluntary process for pre-release model access while it gradually expands access.

Google said it will continue to gather feedback from early testers as it iterates on guardrails before making Argon available to developers, enterprises, and consumers as soon as possible, according to the company blog. The Verge reported that Google will be strengthening critical frontier safeguards before a broader rollout, including defending against misuse and prompt injection attacks and monitoring for misalignment.

According to Ars Technica, general availability will start with paid API users and Google AI Ultra subscribers after cybersecurity tests, although Google is not making specific timeline promises for enterprise and consumer availability. VentureBeat, citing an embargoed Google blog post, reported the same planned order: broad availability is planned as soon as possible for developers, enterprises and consumers, starting with paid API customers and Google AI Ultra subscribers.

Price and output capacity disclosed

Google disclosed introductory pricing for Argon alongside the limited rollout. According to Google, Argon will launch at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off the input token price.

Ars Technica described the same rates as limited-time API pricing: $2 per million input tokens and $10 per million output tokens, with cached input tokens discounted 95 percent.

To support longer, more complex use cases, Google said it is significantly expanding the model's output token limit to 1 million tokens, up from the previous 64,000 tokens. According to Google, the added headroom lets the model think deeply and generate hundreds of thousands of tokens in a single trajectory.

Ars Technica reported the same change, noting the increase from 64,000 tokens in previous Gemini models to a confirmed 1-million-token output limit. The publication reported Google's position that the larger limit allows users to complete more daunting tasks in a single step.

Coding and enterprise work benchmarks

Google positioned Argon around software development, professional knowledge work and cybersecurity operations, areas where VentureBeat noted enterprises are already spending. According to Google, engineers have been using Argon for daily tasks from everyday debugging to large-scale codebase migrations and algorithm designs.

On DeepSWE v1.1, which Google describes as measuring performance in real-world long-horizon software engineering tasks, Argon sets a new state of the art at 77.9%, according to Google. Ars Technica reported the same 77.9% result and said it was higher than GPT-6 Astra, Fable 5.1 and Opus 5.5 in Google's comparison.

Beyond coding, Google said Argon is the leading model on the Vals Index, which measures economic impact across finance, coding, legal and tax work, with every sector weighted by its contribution to U.S. GDP. The company also cited leading performance on Vals Finance Agent v2 for multi-step financial research and Harvey's Legal Agent Benchmark for legal research and drafting, according to the Google blog.

On AutomationBench, Zapier's benchmark measuring end-to-end execution across core business functions, Argon ranks No. 1 with a score of 51.3%, according to Google. For visual work, Google cited LVBench for long video understanding, where it said Argon is state of the art with a score of 91.7% and can drive professional chart analysis, identify details from long videos and take action based on a series of documents.

VentureBeat analyzed the 18 benchmarks Google disclosed and reported that Argon leads outright on 12 and ties for first on one. In that same comparison set, GPT-6 Astra leads outright on three and ties Argon on one, while Claude Opus 5.5 leads outright on two, according to VentureBeat. That tally made Argon the leader by total benchmark leads or top scores in Google's comparison set, although VentureBeat noted OpenAI and Anthropic retained advantages in several technical categories.

Internal use at Google

Google said Argon is already powering internal workflows, with thousands of Googlers highlighting strengths in specialized coding tasks, deeper research and writing quality, according to the company blog. Tulsee Doshi, head of Gemini products at Google DeepMind, told Axios that Googlers had put the model through its paces in recent weeks, with many relying on it for hardest coding and research problems.

Axios quoted Doshi describing Argon as "a well-rounded model that has frontier capabilities across several domains."

Google cited quantum algorithmic optimization as one internal example. According to Google, Argon helped quantum computing researchers optimize spacetime resources, defined as qubits multiplied by gates, for subroutines that bottleneck important applications. In one example, it beat the published baseline by 40% in a matter of minutes.

A second example involved memory efficiency. According to Google, a team of Argon agents analyzed fleet-wide profiling telemetry to autonomously identify and apply memory optimizations across Google's data centers, freeing up over 300 TiB of memory once rolled out, with an estimated 500 TiB to 1 PiB in total savings. Ars Technica reported the same account, saying Argon used fleet-wide telemetry data to help save 300 TiB across data centers.

A third area was large-scale codebase migration. According to Google, Argon agents are working on migrating C and C++ codebases to Rust across Google, scaling from tens of thousands of lines in core libraries like re2 and libgav1 up to more than 800,000 lines for the Fuchsia Zircon kernel. Ars Technica reported the same scope, including thousands of lines in re2 and libgav1 and more than 800,000 lines in the Fuchsia OS Zircon kernel.

Google said such large-scale rewrites are undergoing rigorous automated and manual auditing, emulation testing and review before rolling out to production because of the criticality of many systems. For libgav1, Google's open-source software for decoding video, Argon agents took an existing Rust port and replaced 32,000 lines of SIMD code through profile-guided experiments and compiler-output study, producing safe Rust so the compiler would vectorize it automatically, according to Google. The company said the result is a memory-safe video decoder that runs 2.7 times faster than the Rust port, with identical video output.

Cybersecurity defense focus

Google said it trained Argon to be highly capable at cybersecurity defense and that Argon can autonomously find, validate and patch critical software vulnerabilities. For trusted defenders and Google's own internal teams, Google said it will release Argon without cyber guardrails so they can use its full frontier-level defense capabilities.

Wiz is already using Argon for cybersecurity defense through its Scan for Good initiative, a program dedicated to protecting critical public infrastructure for free by finding and remediating high-risk exposures, according to Google. In an early demonstration, Google said the model uncovered a critical vulnerability exposing sensitive personal information.

Ars Technica reported the same Wiz deployment, saying Wiz used Argon to uncover a critical vulnerability that could expose personal information in a system used at hospitals around the world. Ars Technica added that Google claimed other frontier models missed the flaw but did not provide specifics.

The Verge reported that Google is limiting access at first to trusted cyber defenders in part so the company can make sure the model is not misaligned. Ars Technica noted concern about model misalignment after high-profile hacking incidents over the summer and reported Google's claim that it designed Argon with systems that monitor chain-of-thought and can stop the model if it steps out of bounds.

Context of delays and rival launches

The launch followed months of delay at the top of Google's lineup, according to Axios. Axios reported that Gemini 4 arrives after a long gap in which Google kept rolling out smaller, cheaper Flash models while OpenAI and Anthropic moved forward. Axios also reported that Gemini 4 comes nearly a year after the previous flagship generation, Gemini 3.

In May, Google CEO Sundar Pichai said the next major model release would come in June, according to Axios. That model, widely expected to be Gemini 3.5 Pro, never came out, Axios reported. Ars Technica similarly reported that Google promised Gemini 3.5 Pro in June but spent the summer releasing smaller Flash models.

Axios reported in July that poor morale among Google DeepMind employees was part of the reason for the delay. On Sept. 30, Axios added that Bloomberg, citing anonymous sources, reported some employees found model performance lacking in internal testing, while Google told Bloomberg that characterization was inaccurate.

The timing also placed Argon one day after OpenAI's DevDay conference, according to The Verge. The Verge reported that OpenAI launched its Dots AI agent and GPT-6.1 Sol AI model at that event and said OpenAI would not release a planned GPT-6.1 Astra model because of safety worries. Axios separately noted OpenAI scrapping GPT-6.1 Astra because of security issues and said OpenAI and Anthropic are investigating tens of thousands of incidents involving agents misbehaving.

Sources referenced