October 13, 2026
Gemini 4 Argon now fits a million tokens in a single response
The introductory price holds until Nov 30, 2026: $2 per million input tokens and $10 per million output tokens.
A single Argon response can stretch to 1,048,576 tokens. The previous Gemini 3.1 Pro Preview topped out at 64,000 output tokens.
Long work no longer has to be chopped into pieces by hand: the model finishes a big review or edit within one conversation. The million tokens are a shared pool for input, thinking and output.
Per request, streaming returns at most 65,536 tokens, non-streaming REST 32,768. When a response hits the limit, the API returns finish_reason=CONTINUATION and an encrypted continuation_token: send it again together with the original prompt, system instruction and tools until finish_reason=STOP arrives. The Google Gen AI SDK runs this loop itself when max_output_tokens is above 65,536.
A price on a timer. Until Nov 30, 2026, input costs $2 per million tokens and output $10. From Dec 1, Google raises it to $4 and $20. Cached input is 95% cheaper.
Where to turn it on. Access has been open since Oct 13, 2026: the paid Gemini API tier, Google AI Studio and Android Studio. Google AI Ultra subscribers and eligible Gemini Enterprise plans got Google Antigravity, a desktop app for macOS, Windows and Linux plus a CLI.
In the Gemini app and AI Mode in Search, Argon is enabled for Ultra and Pro subscribers. In the API the model is called gemini-4-argon; thinking depth is set by the thinking_level parameter with values LOW, MEDIUM and HIGH, with MEDIUM as the default.
An outside measurement. Artificial Analysis gave Argon 53 points on its Intelligence Index, level with GPT-6 Astra at its maximum setting. Argon hallucinates 15% of the time versus 51% for Astra.
It uses more tokens: Argon spends 62,000 output tokens per task, Astra 27,000. Even so, a full run of the index came out cheaper, $1.99 versus $3.26. On the agentic AutomationBench, Argon took first place with 78%.
From Dec 1, 2026, the introductory price ends, and a million input tokens will cost $4.