Google released Gemini 4 Argon worldwide on Tuesday, October 13, 2026, pricing its flagship AI model at an introductory $2 per million input tokens. The company says Argon now ranks first on the Vals Index, a benchmark covering finance, coding, legal and tax work.
What Happened?
- Gemini 4 Argon is available now to developers, enterprises and Google AI Pro and Ultra subscribers worldwide.
- Argon scored 68.90% on the Vals Index, ahead of Claude Opus 5.5 at 66.97% and GPT-6 Astra at 63.13%.
- Google’s launch rates double after November 30, and cached input tokens cost 95% less than standard input.
- Google is widening its Fairwind Program, which gives vetted cyber defenders Argon without cyber guardrails, to hundreds more partners.
Argon arrives cheaper than its closest rivals
Argon follows a rough stretch for Google’s AI program. In September, Google dropped plans to ship Gemini 3.5 Pro, which CEO Sundar Pichai had said would arrive in June. The company also overhauled DeepMind after co-founder and CEO Demis Hassabis stepped aside and several Gemini model leaders left. The release puts a new flagship at the top of Google’s Gemini AI model lineup.
Price is where Google pushes hardest. Argon’s introductory output rate is $10 per million tokens, and both rates move to $4 and $20 once the launch window closes. GPT-6 Astra and Claude Fable 5.1 both list at $10 per million input tokens and $50 per million output tokens.
On long-context work, the gap with Anthropic’s models opens up. Argon scored 99.7% on GraphWalks, against 98.7% for GPT-6 Astra, 91.4% for Claude Fable 5.1 and 90.6% for Claude Opus 5.5.
Google’s engineering claims are self-reported but specific. Teams using Argon in Antigravity raised agentic code submissions by more than 35% and cut code rollbacks by nearly a third, the company said. On the open Gemma 4 E2B model, Argon helped deliver a 2.85x inference speedup on an A100 via vLLM in a few days. Throughput climbed from about 37,900 to over 108,000 tokens per second, Google said in its launch post.
Access depends on the plan:
- Developers get the Gemini API paid tier, with Antigravity, AI Studio and Android Studio for Google AI Ultra subscribers.
- Enterprises get Gemini for business, the API and Antigravity.
- Consumers on Google AI Pro or Ultra get Argon in the Gemini app and AI Mode in Search.
- Scientists can apply for 10,000 free six-month seats on Antigravity’s highest usage tier.
Deloitte, LexisNexis, Ambience Healthcare and Isomorphic Labs tested the model before launch.
Cyber defenders get Argon without guardrails
The cyber release carries the most risk. Google’s Fairwind Program gives vetted defenders a version of Argon with its cyber guardrails removed. After a September rollout to select partners, Google is now opening it to hundreds more. Koray Kavukcuoglu, Google DeepMind SVP and Google’s chief AI architect, said trusted defenders would get Argon “without cyber guardrails so they can leverage its full frontier-level cybersecurity defense capabilities.“
The benchmark case for that access is narrow. On CWE-bench v1, which tests whether a model can fix security vulnerabilities, Argon tied at 68% with Grok 4.7 and GPT-6 Astra. Claude Opus 5.5 scored 67%.
Wiz Did the Testing for Argon
Google’s showcase evidence comes from Wiz, a company Alphabet finished acquiring in March 2026. Google says Wiz used Argon to chain healthcare infrastructure flaws, surfacing a public file download bypass that exposed a master key and admin access. The fix protected hundreds of thousands of patient records, Google added.
The announcement doesn’t name the healthcare operator or list a CVE identifier for the bypass. It also doesn’t explain how Google vets Fairwind partners or audits their use of an unguarded model.
External bio red teaming unlocked less than 1% of the bio sub-tasks Google tests for, the company said. On Gray Swan’s Indirect Prompt Injection benchmark, attack success fell to 0.7% across 15 adaptive attempts. Shortcut attempts on coding benchmarks dropped to 2%. Pichai also recently signed a voluntary White House Accord on Super Intelligence alongside OpenAI, Anthropic, Meta, xAI and Nvidia.
What’s Next?
Teams already building on the Gemini API should cost long agentic jobs at post-launch rates now. The same workload bills at double the price after November 30. Google’s continuation protocol documentation covers runs that use the 1 million token unified input-output limit. That price step is the next fixed marker, and Google says the rest of the Gemini 4 family will follow.
