Gemini 4 Argon Doubles Its Price December 1 and Has No Explicit Caching, So Rebuild for Implicit Cache Hits Now

Gemini 4 Argon's introductory pricing ends November 30, explicit caching is not supported, and max_output_tokens now caps thinking and output together. Four changes to make this month before the bill doubles.

Gemini 4 Argon is still on introductory pricing. Through November 30, 2026, the API charges $2.00 per million input tokens and $10.00 per million output tokens. From December 1, those rates rise to $4.00 and $20.00. Cached input moves from $0.10 to $0.20 per million tokens on the same date. Any workload you migrated in the past few weeks has a cost baseline that is about to become wrong.

The bigger trap is caching. Google's documentation says explicit caching, the cachedContents and caches.create path many teams built their long context pipelines around, is not supported on Gemini 4 Argon. What you get instead is implicit caching, which is enabled by default for prompts of at least 4,096 tokens. In practice, that means the savings you planned to engineer by hand now depend on how consistently your requests repeat the same long context, and whether each request clears the 4,096 token threshold at all.

The second trap is output length. On Gemini 4 Argon, max_output_tokens is a hard limit on thinking and output together. A response stops once the two combined reach the cap, so a budget that comfortably fit the visible answer on an older model can now cut the answer off after the model spends the allowance reasoning. Google advises existing apps to raise max_output_tokens to at least 32,768. The whole exchange lives inside one unified 1,048,576 token window shared by input, thinking and output.

Two cleanup items round it out. thinking_level accepts low, medium and high, with medium as the default, so if you never set it you are already paying for medium. And temperature, top_p and top_k are ignored if you send them, which means any tuning you believe those parameters are doing is not happening.

The practical sequence is short: remove the dead code, fix the output cap, measure what implicit caching actually saves you, then rerun the cost model at the December rates. Doing it this week means the December invoice is a number you already know.

Try it today

1. Search your codebase for cachedContents and caches.create calls on Gemini 4 Argon and remove them; explicit caching is not supported on this model.
2. Check that the long, repeated portion of each request keeps the prompt at or above 4,096 tokens, the threshold where implicit caching turns on by default.
3. Raise max_output_tokens to at least 32,768, because on Gemini 4 Argon the limit covers thinking and output together.
4. Set thinking_level explicitly to low, medium or high instead of inheriting the medium default, and test whether low meets your quality bar.
5. Delete temperature, top_p and top_k from Gemini 4 Argon requests, then recalculate monthly cost at $4.00 input, $20.00 output and $0.20 cached input per million tokens.