Agent Upstart

Cool things people are building with agents — and practical help getting the most out of them.

Playbooks · Coding tasks

Gemini 4 Argon doubles in price on December 1

Written November 4, 2026

The introductory price was never the price. It was a dated rate with a second rate printed beside it, and the second rate takes over on December 1, 2026. Any budget, forecast, or per-task cost built on the first rate doubles overnight unless the work moves.

What changes, and when

Gemini 4 Argon (gemini-4-argon) became generally available on the Gemini API on October 13, 2026. Google's Gemini API pricing page lists it as the flagship reasoning model, built for complex software engineering, long agent work spanning many steps, and generation with outputs up to 1M tokens, with a single flat rate across the whole context window. The same page prints two prices for every rate, with a date between them.

At the Standard tier, on the paid tier, per million tokens in US dollars: input is $2.00 through November 30, 2026, and $4.00 starting December 1, 2026. Output, including thinking tokens, is $10.00 through November 30 and $20.00 from December 1. Context caching is $0.10 through November 30 and $0.20 from December 1, and caching is implicit only: explicit caching is not supported, so there is no cache object to create and no cache write charge to budget. The doubling applies at every tier, not only Standard. Batch and Flex run at half the Standard rates in both periods: $1.00 input and $5.00 output through November 30, $2.00 and $10.00 from December 1, with caching at $0.05 rising to $0.10. Priority runs at $3.60 input and $18.00 output through November 30, $7.20 and $36.00 from December 1, with caching at $0.18 rising to $0.36.

There is no free tier for Argon. The pricing page states plainly that the model is not available on the Free tier, and that every token the model processes is counted and billed, including thinking tokens and internal tool system instructions and function declarations when grounding is used. A task that reads 100,000 uncached input tokens and writes 10,000 output tokens costs about $0.30 at the introductory Standard rates and about $0.60 at the December 1 rates, before tools, taxes, and retries. The arithmetic is exact at list prices; the token counts in real agent work are not, because thinking tokens are billed as output and are not visible in the answer.

Where Argon is reached

Argon is reached through the Gemini API on the paid tier, in AI Studio with the model id gemini-4-argon. The pricing page is an API pricing page: it says nothing about what a subscription plan, a coding assistant, or a third-party gateway charges for the same model, and those routes can carry different prices, different included usage, and different rollout dates. A builder using Argon through GitHub Copilot, Antigravity, Android Studio, or any other host should check that host's own price and availability wording rather than assuming the API list price applies there. If a host shows Argon's introductory price, the question to ask it is the same one the pricing page answers for the API: what rate applies from December 1, and does the host pass the change through on that date.

Grounding is a separate charge on the Standard and Priority tiers: 5,000 free search requests per month, shared across all Gemini 3 and newer models, then $14 per 1,000 requests, subject to a daily cap of 1,500 requests per day. Batch and Flex do not list grounding. A task that grounds every step pays for searches on top of tokens, and the search charge does not double on December 1 — only the token rates do.

What to do before December 1

  1. Reprice one real workload at both rates. Take a week's actual Argon usage — input tokens, output tokens including thinking, and cached input — and price it at the introductory rates and at the December 1 rates from the pricing page. The gap between the two totals is the decision, not the headline doubling: a workload dominated by cached input doubles from a small base, while a workload dominated by fresh input and long thinking doubles from a large one.
  2. Decide whether Batch or Flex fits the work. Both run at half the Standard rates in both periods. Work that can wait for batch processing, or that tolerates flexible scheduling, costs half as much on December 1 as the same work at Standard — $2.00 and $10.00 instead of $4.00 and $20.00. Work that cannot wait stays at Standard, and the December 1 rate is the rate to budget.
  3. Check the route, not only the model. For each place Argon is used — API key, AI Studio, a coding assistant, a gateway — confirm the rate that applies from December 1 in that route's own terms. A route that absorbs the change, delays it, or adds its own margin will not match the pricing page, and the pricing page cannot settle what a host charges.
  4. Measure thinking separately. Output is billed including thinking tokens, so a task whose reasoning is long and whose answer is short is mostly paying for tokens nobody sees. Log input, output, and cached tokens per task before December 1, so the doubled bill can be traced to the token class that caused it rather than to the model in general.
  5. Do not build a December budget on the November rate. Any forecast, quote, or per-task cost that uses $2.00 and $10.00 without a date is wrong from December 1. State the rate period beside every Argon figure, in the forecast and in any message to a client or teammate who will pay it.

What can go wrong

Limits

Related playbooks

Sources

Back to: Coding playbooks hub · Also: Use your own Anthropic API key with an agent: what it costs, and the billing trap to check first · What an agent costs, in plain figures · Agent cost estimator · Move off Claude Sonnet 4.5 before it retires