Gemini 4 Argon prices double Dec 1: $4 input, $20 output, thinking included

Express-Collection11 · reddit · 2026-10-22

Gemini 4 Argon went GA on Oct 13, but the current pricing is temporary: $2/M input and $10/M output only through Nov 30, doubling to $4 and $20 on Dec 1. Output billing includes thinking tokens—where reasoning traffic lives—implicit cache hits go from $0.10 to $0.20, and Argon has no free tier.

Migration notes:

Context window is 1,048,576 tokens shared across input, thinking and output; maxoutputtokens (default 65,536) also covers thinking, and Google advises raising limits under 32k to at least 32,768.
thinkinglevel: low/medium (default)/high—Google recommends evaluating all three and picking the lowest that meets quality, the real cost lever at this pricing.
temperature/topp/topk are silently ignored (unlike Sonnet 5.5 which rejects them); explicit caching is unsupported.
On the Interactions API, outputs above 64k need background: true and arrive in 64k hops; Google recommends client timeouts of at least 15 minutes.

Practical move: measure cost per request at each thinking level while it's cheap, then double estimates for December; track cost/tokens per call from day one via gateways like Litellm's per-key budgets.

Original post →

Gemini 4 Argon prices double Dec 1: $4 input, $20 output, thinking included

More from Models