Alphabet Inc. introduced Gemini 4 Argon on Wednesday, September 30, 2026, positioning it as its most advanced artificial intelligence model to date.
The model is designed to enhance coding capabilities, cybersecurity defenses, and complex professional workflows.
Google is rolling out Gemini 4 Argon in phases, initially granting access to select cybersecurity partners.
The company is also participating in pre-release safety evaluations with the United States government.
This deployment aims to safeguard against misuse, prompt injection attacks, and system misalignments before a broader public launch.
Performance and Capabilities
Gemini 4 Argon expands output capacity to one million tokens and delivers top performance on industry benchmarks.
It scored 77.9 percent on DeepSWE v1.1 for software engineering and leads specialized evaluations in legal and financial research.
The model also ties for top performance in cybersecurity vulnerability remediation.
Internally, Google has deployed the model to automate memory optimization across its data centers, freeing over 300 tebibytes of memory without additional hardware purchases.
Development teams are using Argon agents to execute large-scale codebase migrations from C and C++ to Rust.
The announcement follows a voluntary AI safety agreement signed at the White House by CEO Sundar Pichai alongside President Donald Trump and major technology executives.
The model will initially be available to Google AI Ultra subscribers and paid API customers at an introductory rate of $2 per million input tokens and $10 per million output tokens.
Despite benchmark successes, people with direct access to the development effort noted internal skepticism regarding the model's performance on certain practical coding tasks.
The new flagship model is larger than previous top-tier Pro models, following the cancellation of Gemini 3.5 Pro as Google pivoted toward frontier intelligence capabilities.
"Starting this rollout in this way gives us more confidence, but also enables us to put a model that is trained and strong in cyber defense in the hands of defenders as soon as possible," said Tulsee Doshi, product lead for Google's Gemini models.
Doshi highlighted the system's ability to execute long, multi-step tasks while providing significant improvements over previous iterations like the 3.8 Flash Cyber model.
"We really believe that a model of this caliber and this level of frontier performance is meaningfully important for defenders," Doshi said.
Company officials emphasized that the new architecture addresses both defensive needs and complex corporate operations as competition intensifies among frontier AI developers.
"It's our most performant model yet built for complex workloads, and we see it comparable to frontier models like (OpenAI's) Astra and (Anthropic's) Opus on key coding and cyber benchmarks," said a company spokesperson.
