Google may have taken its time, but seems to have delivered a resounding response in the race for the most advanced artificial intelligence (AI; now called Super Intelligence, or SI) models yet. The Gemini 4 Argon model is now in the hands of what Google calls ‘a set of trusted cyber defenders’, and is working with the U.S. government’s voluntary process for pre-release model access. That means, not everyone can access Gemini 4 Argon just yet. This is Google’s response to OpenAI GPT-6 Astra and Claude Fable 5.1 models.
“To support Gemini 4 Argon’s capabilities across longer, more complex use cases, we are significantly expanding the model’s output token limit to an industry-leading 1M tokens, up from the previous 64K tokens. When the model has the headroom to think deeply and generate hundreds of thousands of tokens in a single trajectory, it adds a new level of depth in reasoning to solve tough problems in one go,” says Koray Kavukcuoglu, SVP, Google DeepMind and Chief AI Architect, Google.
The early benchmark scores give Gemini 4 Argon a distinct advantage over GPT-6 Astra and Claude Fable 5.1 as well as the Claude Opus 5.5 models. In the Vals benchmark which tries to measure AI’s potential economic impact by weighting agentic model performance across finance, coding, legal, and tax tasks, weighted by each sector’s share of U.S. GDP., Gemini 4 Argon returns an accuracy of 68.90%, ahead of Claude Opus 5.5 (66.97%), GPT-6 Astra (63.13%) as well as Meta Muse Spark 1.3 Max (58.16%) and SpaceXAI’s Grok 4.7 (54.95%).
Gemini 4 Argon seems to be consistently leading across the agentic coding, science and math compute, long context, computer tasks and multimodal understanding as well. Google’s bet is that Gemini 4 Argon’s capabilities with coding, reasoning and multimodality will allow it to sustain complex multi-step tasks that would normally define an enterprise workflow.
{{/usCountry}}Gemini 4 Argon seems to be consistently leading across the agentic coding, science and math compute, long context, computer tasks and multimodal understanding as well. Google’s bet is that Gemini 4 Argon’s capabilities with coding, reasoning and multimodality will allow it to sustain complex multi-step tasks that would normally define an enterprise workflow.
{{/usCountry}}Cybersecurity is a big focus area for Gemini 4 Argon, and it seems to be ahead of its closest competition in the CWE-bench v1 benchmark that evaluates a model’s ability to remediate security vulnerabilities.
This is Alphabet’s most advanced model yet, and comes days after CEO Sundar Pichai signed a voluntary White House Accord on Super Intelligence with other AI leaders. Anthropic, OpenAI, Google, Meta, xAI and Nvidia pledged to implement stronger controls as well as monitoring and detection mechanisms internally—this comes after Anthropic’s Dario Amodei raised an alarm about humans losing control of AI, and OpenAI’s models exiting their testing environment and accessed platforms on the open internet.
“To better equip cyber defenders for the new era of cyberattacks, we trained Gemini 4 Argon to be highly capable at cybersecurity defense. Argon can autonomously find, validate, and patch critical software vulnerabilities. For trusted defenders and our own internal teams at Google, we’ll be releasing Argon without cyber guardrails so they can leverage its full frontier-level cybersecurity defense capabilities,” says Kavukcuoglu.
In the CWE-bench v1 benchmark leaderboard, Gemini 4 Argon ties at the top with a score of 68%, and ties with Grok 4.7 as well as GPT-6 Astra, but remains ahead of Claude Opus 5.5 (67%), Claude Fable 5.1 (58%) and Muse Spark 1.3 (55%).
The big question is, when does everyone get access to Gemini 4 Argon? Google says it is imperative to strengthen safeguards in four key areas—defending against misuse, against prompt injection attacks, monitoring for misalignment and hardening the sandboxing environments.
“We’re grateful for the initial cohort of cyber defenders and trusted testers whose real-world evaluations and feedback will help us strengthen our systems before we release to developers, enterprises, and consumers, starting with paid API customers and Google AI Ultra subscribers,” Kavukcuoglu explains.
Speaking of costs, and in a rare turn of events which sees an AI company talk about token costs, Google says 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 input token price. Post the introductory pricing period, that’s expected to be $4 and $20 respectively, per million tokens. In comparison, GPT-6 Astra and Claude Fable 5.1 are priced at $10 and $50 per million input and output tokens respectively.