TL;DR: Google on Sept. 30, 2026 unveiled Gemini 4 Argon and is first offering it only to trusted cyber defenders while it completes U.S. voluntary pre-release safety checks.

Google names Argon as first Gemini 4 model

Alphabet's Google on Wednesday, Sept. 30, 2026 announced Gemini 4 Argon as its new top-tier artificial intelligence model to anchor its Gemini 4 generation, according to Reuters.

The company described Argon as larger in size than its previous line of advanced "Pro" models, Reuters reported, citing a company spokesperson. Google's announcement on blog.google, authored by SVP, Google DeepMind and Chief AI Architect Koray Kavukcuoglu, called Argon its next era of frontier intelligence built to sustain deep reasoning across complex, long-horizon workflows.

Google said Argon delivers frontier performance in complex workflows across real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense, according to blog.google. CNBC reported that the company characterized the system as its most advanced model yet, with major improvements in coding, cybersecurity and complex professional work.

The debut arrived nearly a year after Gemini 3 had returned the company to the forefront of the model race, according to CNBC. Axios reported that Gemini 4 arrives after a long gap at the top of Google's lineup, during which the company kept releasing smaller, cheaper Flash models while rivals OpenAI and Anthropic moved forward.

Limited rollout starts with cyber defenders

Google is not offering Argon broadly at launch. The model is rolling out to a set of trusted cyber defenders through its Fairwind Program, according to blog.google.

The Verge reported that Google is limiting access at first to that defender group so it can confirm the system is not misaligned. Google said safely releasing frontier capabilities at this level requires a phased approach, according to blog.google.

The company said it is actively engaged in the U.S. government's voluntary process for pre-release model access while it gradually expands access, according to blog.google. CNBC reported that Google plans to launch in phases, starting with trusted cybersecurity partners while working with the U.S. government on pre-release safety evaluations.

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 blog.google. Axios reported that paying subscribers will be first in line after additional testing.

Safety checks gate wider availability

Google said it will strengthen critical frontier safeguards before a broader rollout, including defending against misuse and prompt injection attacks and monitoring for misalignment, according to The Verge.

CNBC reported that Google is looking to scale safeguards in four key cybersecurity areas, including misuse and prompt injection, before launching publicly. The phased cyber-defender start is intended to build confidence and place a defense-oriented model with defenders quickly.

Tulsee Doshi, Google's Gemini model product lead, told CNBC that starting the rollout in this way gives the company more confidence and enables it to put a model trained for cyber defense into defenders' hands as soon as possible. Doshi described Gemini 4 as an incredibly well-rounded model that excels at running long, multi-step tasks, according to CNBC.

Doshi told Axios that Argon is a well-rounded model with frontier capabilities across several domains and that Googlers had put it through intensive use in recent weeks for difficult coding and research problems. CNBC quoted Doshi saying the company believes frontier performance of this caliber is meaningfully important for defenders.

Output limit jumps to 1 million tokens

To support longer and more complex use cases, Google is expanding Argon's output token limit to 1 million tokens, up from the previous 64,000 tokens, according to blog.google.

Google said the headroom to think deeply and generate hundreds of thousands of tokens in a single trajectory adds depth in reasoning for difficult problems in one attempt, according to blog.google. Engadget reported that the 1-million-token output limit is several times higher than GPT-6 Astra's 128,000 tokens.

CNBC reported on Oct. 1, 2026 that a token is equivalent to about three-quarters of one word. That definition places the expanded output ceiling in the range of hundreds of thousands of words of continuous generation and reasoning in a single run.

Internal Google use spans quantum, memory and Rust

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 blog.google.

For quantum algorithmic optimization, Google said Argon is helping quantum computing researchers optimize spacetime resources, defined as qubits multiplied by gates, for subroutines that bottleneck important applications, according to blog.google. In one example, it beat the published baseline by 40% in a matter of minutes. CNBC separately reported that quantum computing researchers have utilized the model.

For memory efficiency, Google said a team of Argon agents analyzed fleet-wide profiling telemetry to autonomously identify and apply memory optimizations across Google's data centers, according to blog.google. The company said the work freed up over 300 TiB of memory once rolled out, with an estimated 500 TiB to 1 PiB in total savings. CNBC reported that the effort freed up hundreds of terabytes of memory without buying additional hardware, while Engadget reported the same 300 TiB figure.

For large-scale codebase migrations, Google said Argon agents are working on migrating C and C++ codebases to Rust, 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, according to blog.google. Google said such large-scale rewrites of critical systems are undergoing rigorous automated and manual auditing, emulation testing and review before rolling out to production.

Google said that for libgav1, its open-source software for decoding video, Argon agents took an existing Rust port and replaced 32,000 lines of SIMD code through many rounds of profile-guided experiments and study of compiler output, according to blog.google. The company said the agents produced safe Rust so the compiler would vectorize it automatically, yielding a memory-safe video decoder that runs 2.7 times faster than the Rust port with identical video output.

Benchmarks claim leads in coding and office work

Google engineers have used Argon for daily tasks from debugging to large-scale migrations and algorithm design, according to blog.google. The company said it sets a new state of the art on DeepSWE v1.1 at 77.9%, which measures performance in real-world long-horizon software engineering tasks.

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, according to blog.google. The company reported similarly leading performance on domain evaluations including Vals Finance Agent v2 for multi-step financial research and Harvey's Legal Agent Benchmark for legal research and drafting.

On AutomationBench, Zapier's benchmark measuring end-to-end execution across core business functions, Google said Argon ranks No. 1 with a score of 51.3%, according to blog.google. CNBC summarized the results as a new record in real-world software engineering and a lead on a benchmark covering finance, legal and other professional tasks.

For visual understanding, Google said Argon can drive professional chart analysis, identify details from long videos and take action based on a series of documents, according to blog.google. On LVBench, which measures long video understanding, Google reported state-of-the-art performance with a score of 91.7%.

Independent benchmarking company Artificial Analysis found that Gemini 4 Argon matches GPT-6 Astra's Intelligence Index composite score at 60% of the cost per task at current discounted prices, according to Engadget. Engadget reported that Argon scored one point ahead of OpenAI's GPT-6.1 Sol and had a hallucination rate of 15%, described as the lowest among leading models, compared with 54% for GPT-6 Astra and 54% for GPT-6.1 Sol.

Cyber defense focus shapes first release

Google said it trained Argon to be highly capable at cybersecurity defense so cyber defenders are better equipped for a new era of attacks, according to blog.google. The company said 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, according to blog.google. Wiz is already using Argon for 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 blog.google.

CNBC reported that Argon ties with OpenAI's GPT-6 Astra and Grok 4.7 on cybersecurity benchmarks and sits ahead of GPT-6 Astra and recent Anthropic models on the Vals Index. The outlet also reported that the new model represents a step up from Google's 3.8 Flash Cyber model released earlier in September 2026, outperforming it in vulnerability discovery.

Reuters reported that Google sees Argon as comparable to frontier models like OpenAI's Astra and Anthropic's Opus on key coding benchmarks. Engadget reported that a Google spokesperson described Argon as comparable to OpenAI's GPT-6 Astra and Anthropic's Opus on key benchmarks.

Price, rivals and timing frame the launch

Google said 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, according to blog.google. CNBC reported on Oct. 1, 2026 that the introductory pricing matches OpenAI's newly discounted GPT-6.1 Sol model.

Engadget reported that Astra costs $10 per million input tokens and $50 per million output tokens, making Argon's introductory rates substantially lower before any future change. CNBC reported that Google said its published Argon rates will eventually double to $4 per million input tokens and higher output pricing.

The Verge reported that the launch came just a day after OpenAI's DevDay conference, where OpenAI launched its Dots AI agent and GPT-6.1 Sol model. The outlet also reported that OpenAI said this week that it would not release a planned GPT-6.1 Astra model because of safety worries.

The model release came a day after CEO Sundar Pichai signed a voluntary agreement with President Donald Trump and major tech executives at the White House, following a luncheon to address rising AI safety concerns, according to CNBC. CNBC reported that Pichai signed the voluntary AI safety accord alongside major tech leaders that week.

Axios reported that Gemini 4 was widely expected to follow a different path, noting that Pichai said in May that the next major release would come in June, that the widely expected Gemini 3.5 Pro never appeared, and that Axios reported in July that poor morale among Google DeepMind employees was part of the reason for delay. Axios also reported that Bloomberg, citing anonymous sources, said some employees found the model's performance lacking in internal testing, and that Google told Bloomberg that characterization was inaccurate. The Verge noted that the launch follows Kavukcuoglu's appointment as the new head of DeepMind in August.

Sources referenced