Google launched Gemini 4 Argon, its latest frontier model, on October 13, 2026. In an announcement by Koray Kavukcuoglu, SVP of Google DeepMind and Chief AI Architect, the company positioned the model for deep work on complex tasks in coding, finance, legal, cybersecurity, and science. Read the official announcement.
Availability and pricing
Gemini 4 Argon is available globally from launch day. Developers can build with the Gemini API paid tier in AI Studio and Android Studio, with documentation covering a new continuation protocol and a 1M unified input-output token limit. Enterprises get the model in Gemini for business, the API, and Google Antigravity. Consumers can use it in the Gemini app and AI Mode in Search with a Google AI Ultra or Pro subscription, and in Google Antigravity and AI Studio with an Ultra subscription.
The model launches at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95% off. After the introductory period ends on November 30, 2026, the public-access Argon model moves to $4 per million input tokens and $20 per million output tokens.
Benchmarks
Google reports that Argon ranks first on the Vals Index, which measures model performance across finance, coding, legal, and tax work. On the Embodied Reasoning (ERQA) benchmark the model scores 87%, which the company says outperforms other frontier models relying on pure reasoning. Cited safety results include a 0.7% attack success rate on the Gray Swan Indirect Prompt Injection benchmark across 15 adaptive attempts, external bio red teams unlocking less than 1% of tested bio sub-tasks, and shortcut attempts dropping to 2% on coding benchmarks.
Reported deployments
The announcement lists internal and partner results: an architectural refactor in Chromium that completed months of work in weeks, a 40% reduction in in-memory session preparation time for the Agent Development Kit, a 2.3x speedup for the xTB quantum chemistry package, and a 2.85x inference speedup for Gemma 4 E2B through vLLM. Research applications span quantum error correction with AlphaQubit, new counterexamples to Grad's conjecture for stellarator design, rare-disease diagnosis support at Gamow Labs, and laboratory-confirmed enzyme designs. In robotics, a video-to-simulation-to-real pipeline transferred a learned cornhole skill to Apptronik's Apollo humanoid, which sank seven consecutive shots. Google also says the Fairwind security program is expanding to hundreds of additional partners, and that 10,000 scientists will receive Antigravity's highest usage tier free of charge for six months.
What comes next
Google describes Argon as the first member of the Gemini 4 family and says the rest of the family will follow. A separate post covers the model's safety approach, alongside a published model card. Our source verification guide describes how we read official announcements.
