Gemini 4 Argon pricing: tokens, batch, search and Dec 1

Gemini 4 Argon pricing is $2 per million input tokens and $10 per million output tokens through 2026-11-30, then $4 and $20 from 2026-12-01 (Gemini API pricing, read 2026-10-15). Output includes thinking tokens, and there is no free tier.

Google made Argon generally available on 2026-10-13, and its pricing page shows an update dated 2026-10-15. This post lists every Argon price Google publishes today, works out what one agent task costs before and after the change, and shows where search grounding adds to the bill. By the end you can price your own workload in a few lines.

Last checked 2026-10-15, against Google's pricing pages and looot's live catalog.

Gemini 4 Argon pricing, every line Google publishes

Argon has two price periods. The introductory rates run through 2026-11-30 and double on 2026-12-01 for input, output, caching and batch alike. Search grounding does not change: 5,000 free search requests a month across Gemini 3 and newer models, then a flat price per 1,000.

Explicit context caching is not offered for Argon. Caching is implicit: Google applies it to a repeated prompt prefix on its own, per the Gemini 4 Argon model page (read 2026-10-15).

Is Gemini 4 Argon free to try?

No. Google's pricing page says Argon is not on the free tier or the Google AI Pro plan. To call it you need a project with a linked billing account, which is Tier 1 on Google's rate limits page (read 2026-10-15), with a Tier 1 billing cap of $250.

The Gemini app is a separate product. Google's launch post (read 2026-10-15) says Argon is in the app for Google AI Ultra and Pro subscribers, but that plan does not include API calls.

What does one agent task cost on Gemini 4 Argon?

Take a task that reads 20,000 tokens of context, writes 2,000 tokens of thinking and answer, and runs three Google searches. Here is the arithmetic at Google's rates:

Task: 20,000 input tokens, 2,000 output tokens (thinking included), 3 searches

Through 2026-11-30   input   20,000 x 2 / 1M  = 0.040 USD
                     output   2,000 x 10 / 1M = 0.020 USD
                     tokens total              = 0.060 USD
From 2026-12-01      input   20,000 x 4 / 1M  = 0.080 USD
                     output   2,000 x 20 / 1M = 0.040 USD
                     tokens total              = 0.120 USD
Searches, once the free 5,000 are used: 3 x 14 / 1,000 = 0.042 USD

So 1,000 such tasks cost about $60 in tokens now and $120 after the change, plus $42 for searches past the free allowance (rates from Gemini API pricing, read 2026-10-15). The searches can cost as much as the tokens. Google bills each search query the model decides to run, so a prompt that triggers five searches counts five times (Grounding with Google Search docs, read 2026-10-15).

Thinking is the other lever. Argon takes a `thinking_level` of `LOW`, `MEDIUM` (the default) or `HIGH`, according to Google Cloud's Gemini 4 Argon developer guide (read 2026-10-15). Thinking tokens are billed as output, so a lower level on routine steps lowers the output line.

How does Gemini 4 Argon pricing compare with GPT-6.1 Sol and Claude Opus 5.5?

Today Argon and GPT-6.1 Sol cost the same per token. From 2026-12-01, Argon costs the same as Claude Opus 5.5 and twice GPT-6.1 Sol.

The full side by side, with context windows, output limits and search tools, is in Gemini 4 Argon vs GPT-6.1 Sol vs Claude Opus 5.5.

Gemini API pricing on Google Cloud's Agent Platform

Google Cloud sells the same model through its Agent Platform at the same global price. The introductory rate there works as credits back rather than a lower list price.

The page describes the introductory price as "50% credits back on net spend", and it says the promotion does not apply to the Gemini 4 Argon Cyber Permissive variant.

How do you lower a Gemini 4 Argon bill?

Three changes cut the bill without changing the model.

1. Batch what can wait. Batch input and output cost half the standard rate in both periods.
2. Put the fixed part of the prompt first. Implicit caching works on a shared prefix; Google Cloud's guide suggests static content of 6,144 tokens or more at the start of the prompt.
3. Count your searches. Each grounded search is billed. When a step needs Google results with their positions, a page as text or a company record, a data tool can be cheaper than grounding and returns things grounding does not.

Here is what those data jobs cost today, read from looot's live catalog:

The comparison with grounding, at real volumes, is in Grounding with Google Search pricing. Setup for those tools is in Gemini 4 Argon MCP, and the key and first call are in the Gemini 4 Argon API guide.

What it cannot do

• looot does not sell Gemini 4 Argon tokens. Argon is not in looot's catalog today, so you pay Google for the model at the rates above.
• It cannot change Google's prices or the 2026-12-01 date. Google updated its page on the day we read it, so read it again before you budget.
• A data job adds its own price per call, shown live above, and the text it returns counts as Argon input tokens.
• It does not run on a timer. A weekly price check needs scheduled runs, not available yet.