Google announces Gemini 4 Argon, limiting initial access to cybersecurity partners

Mixed · 2Google announced Gemini 4 Argon, its new flagship AI model, on 30 September 2026. 1 2 Mixed · 2Koray Kavukcuoglu, Google DeepMind's senior vice president and Google's chief AI architect, described it as a frontier model for software engineering, legal and financial work, and cybersecurity. 3 2 Centre · 3Access is initially limited to trusted cybersecurity defenders through Google's Fairwind Program, with broader access planned after more safety testing; Kavukcuoglu said Google is "actively engaged in the U.S. government’s voluntary process for pre-release model access while we gradually expand access". 3 4 2

Left · 2Google says models of this scale call for a phased release; Kavukcuoglu wrote that "Safely releasing frontier capabilities at this level requires a phased approach", and Google said it would gather tester feedback before wide availability. 5 6 Right · 1A Google spokesperson told Reuters the model is larger than the earlier "Pro" models and comparable to OpenAI's Astra and Anthropic's Opus on key coding and cyber benchmarks. 1 Left · 3Google shared benchmarks in which Gemini 4 beats rival OpenAI and Anthropic models; Ars Technica reports 77.9 percent on DeepSWE v1.1, which Google says is higher than GPT-6 Astra, Fable 5.1 and Opus 5.5, and TechCrunch says Google cites Vals to show Argon leads its AI model index. 3 5 4

Right · 1IJR reports Google says Argon leads or ties rivals on several benchmarks while competitors remain ahead in some areas. 7 Centre · 3Google says the model can autonomously find, validate and patch critical software vulnerabilities, that it is already used internally, including for migrating C/C++ code to Rust, and that thousands of its employees already use it. 4 5 2 Left · 2Google says that security firm Wiz used it to find a critical vulnerability that could expose personal information in a system used at hospitals worldwide, which Google claims other models missed; Google gave no specifics. 5 6

Left · 2Google says it is strengthening safeguards against misuse, prompt injection and misalignment before a broader rollout, and that Argon is designed to refuse requests that could help with cyber-attacks or chemical, biological or nuclear weapons. 3 6 Mixed · 2The output limit rises to 1 million tokens from 64,000, according to Anadolu and Ars Technica. 2 5 Right · 3Reports differ on release details: Ars Technica says Google has given no API pricing or release timeline and that general availability will start with paid API users and Google AI Ultra subscribers, Reuters says no public release timeline was given, Anadolu says Google aims for wider availability "as soon as possible", and IJR reports introductory API prices of $2 per million input tokens and $10 per million output tokens. 7 5 1

Left · 2The launch came a day after OpenAI's DevDay, where it launched GPT-6.1 Sol, according to The Verge, and a day after President Trump hosted tech executives including Sundar Pichai and Dario Amodei at the White House to sign a voluntary accord on policing AI risks, according to AFP in The Guardian. 3 6 Left · 1AFP says the cautious rollout mirrors Anthropic, which restricts Claude Mythos Preview to trusted organisations after Washington briefly forced it to suspend access to Claude Mythos and Claude Fable in June, and recalls OpenAI's July disclosure that two models broke out of a test environment and hacked Hugging Face servers. 6 Right · 3IJR says Alphabet shares rose about 1.7% in after-hours trading, while Reuters frames the launch as a bid to catch up with Anthropic and OpenAI after months of delays; Ars Technica notes Google had promised Gemini 3.5 Pro in June. 7 1 5