Gemini 4 Argon vs Gemini 3.1 Pro: API Price, Output Limit and When to Switch

The December 1 price change reverses which model is cheaper

Read across the output row: Argon is $2 cheaper than 3.1 Pro on output until November 30 and $8 dearer from December 1 on prompts under 200K. Argon has no long-prompt tier, so the gap narrows on very long inputs, where 3.1 Pro's price already rises.

Two illustrative requests, using these rates (ToolColumn arithmetic):

After December 1, Argon costs nearly twice as much as 3.1 Pro on the shorter request and about the same on the longer one. If most of your calls are short, run Argon's cost through your own token logs before switching everything.

Argon requests: 1M-token long decode, thinking levels, implicit caching only

• Output length. 3.1 Pro returns up to 65,536 tokens. Argon's long decode keeps generating up to 1M tokens minus the input; its per-request cap is 65,536 on the model page, but Google's latest-model guide gives 32,768 for non-streaming requests and 64K for streaming, so stream long answers.
• Thinking controls. Google's guide to Argon uses thinking levels, low, medium (the default) and high. A thinking_budget still works but maps to the nearest level and returns a warning, and the guide tells you to remove temperature, top_p, top_k and candidate_count from requests, so check any 3.1 Pro code that sets them.
• Caching. Argon supports only implicit caching, with no explicit caches and no storage fee. If your 3.1 Pro code creates explicit caches, that code has to change, and you stop paying the hourly storage charge.
• Inputs. Argon takes text, images, video, audio and PDFs and returns text only. Computer use is in preview. Its knowledge cutoff is March 2026; the 3.1 Pro model page gives none.

Google's Argon tables leave 3.1 Pro out

Argon's model card compares it with GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5, not with 3.1 Pro, and Google's Gemini product page adds Gemini 4 Carbon instead. Google's model card calls Argon a new frontier model rather than a fine-tune of an earlier one, and the launch post claims a new state of the art on DeepSWE v1.1 (77.9%) and an output limit raised from the previous 64K tokens. None of those figures has a 3.1 Pro number beside it, so how much better Argon runs your prompts is something only your own logs will show.

In the Gemini app, AI Pro is the cheapest plan with Argon

Google AI Pro ($19.99 a month) and Ultra ($99.99 or $199.99) add Argon next to the Pro model, which Google's plan page still describes as 3.1 Pro. AI Plus ($4.99) gets the Flash model and free accounts Flash-Lite, so neither includes Argon or 3.1 Pro. Google publishes no Argon-specific allowance: app limits refresh every five hours up to a weekly cap, and AI Pro has four times the standard usage. Argon is available in select countries.

For coding, Google's pages disagree about Antigravity: the Argon launch post names AI Ultra, and the Antigravity blog names AI Pro and Ultra. Check the model picker before upgrading for it.