Gemini 4 Argon: Google’s Frontier Model Goes Global

Google just opened the door on its latest frontier stack. On October 13, 2026, Koray Kavukcuoglu announced Gemini 4 Argon as globally available for developers, enterprises, scientists and eligible Gemini subscribers — with a matching regional blog mirror.

Where you can use it

• Developers: Gemini API paid tier in AI Studio and Android Studio; Google AI Ultra also gets Antigravity
• Enterprises: Gemini for business, the API and Antigravity
• Consumers: Gemini app and AI Mode in Search for Google AI Ultra and Pro subscribers
• Researchers: apply for six months of Antigravity’s highest usage tier free (eligibility and capacity caps apply)
• Cyber defenders: Fairwind Program expansion for Argon access without cyber guardrails (application-based)

Pricing and context window

Introductory API pricing is $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95% off. After November 30, 2026, Google says the public Argon SKU moves to $4 / $20. Argon ships with an industry-leading 1M unified input-output token limit on the Gemini API and a documented continuation protocol aimed at long agentic jobs that previously hit hard cutoffs.

What Google claims Argon is good at

Google positions Argon for deep multi-step work in coding, finance, legal, cybersecurity and science, and says it ranks #1 on the Vals Index across those economic domains. The launch post walks through internal examples: a large Chromium architectural refactor that separated core webpage handling from a multi-million-line features folder; overnight ADK Python optimizations that cut in-memory session prep by about 40% and cold imports by 21%; Gemma inference throughput gains on A100 via vLLM; and Fairwind partners such as Wiz using early access on infrastructure hardening. Treat those figures as vendor-reported until independent benches catch up.

Safety framing

Google calls Argon its most comprehensive safety and security stack yet under the Frontier Safety Framework: stronger refusal on harmful requests, a real-time misalignment monitor on agent actions, and a Gray Swan indirect prompt-injection result Google cites at 0.7% attack success across 15 adaptive attempts. Shortcut attempts on coding evals are said to drop to about 2%.

Rocks take

Another frontier model with agentic packaging and a clock on intro pricing. For DIY smart-home builders the interesting bit is less the Vals Index and more whether Antigravity plus the 1M window make Home Assistant / automation coding cheaper and more reliable than Claude or GPT agent SKUs once the November price step hits.