AI Basics & TutorialsGemini 4 ArgonGoogle AIAPI pricingGemini APIAI costs

Gemini 4 Argon Prices Double on December 1: What the $2 to $4 Jump Costs You, and 4 Ways to Respond

Google's Gemini 4 Argon costs $2 per million input tokens and $10 per million output tokens through November 30, 2026. On December 1 both rates double to $4 and $20, so the same workload costs twice as much unless you change tiers, cache more, or switch models before the deadline.

Published November 10, 2026

Muhammad Usman

By Muhammad Usman · Founder & Lead Reviewer

Gemini 4 Argon Prices Double on December 1: What the $2 to $4 Jump Costs You, and 4 Ways to Respond

Quick Answer

Google's Gemini 4 Argon costs $2 per million input tokens and $10 per million output tokens through November 30, 2026. On December 1 both rates double to $4 and $20, so the same workload costs twice as much unless you change tiers, cache more, or switch models before the deadline.

Key takeaways:

  • Gemini 4 Argon launched October 13, 2026 at an introductory $2 input and $10 output per million tokens.
  • Every rate doubles on December 1, 2026: Standard $2/$10 becomes $4/$20, Batch and Flex $1/$5 become $2/$10, Priority $3.60/$18 becomes $7.20/$36, and cached input $0.10 becomes $0.20.
  • Independent per-task testing already found Argon more expensive per task than Claude Opus 5.5 at the intro price, $1.99 against $1.82, so the December doubling widens that gap.
  • Argon has no free tier, and its output price includes thinking tokens, so reasoning-heavy work pays the output rate for tokens you never see.
  • Four responses before November 30: move batchable work to Batch or Flex, use context caching, measure your real usage, and compare GPT-6.1 Sol at $2/$10 before you commit.

What exactly doubles on December 1?

Google put Gemini 4 Argon on general availability on October 13, 2026, with an introductory price of $2 per million input tokens and $10 per million output tokens. The launch post carried the second price in a footnote: after the introductory period expires on November 30, 2026, the price becomes $4 per million input tokens and $20 per million output tokens. Google's own Gemini API pricing page lists both sets of rates, dated through November 30 and from December 1.

The doubling covers every tier, not just Standard. Batch and Flex, which run at half the Standard price, move from $1 input and $5 output to $2 and $10. Priority moves from $3.60 and $18 to $7.20 and $36. Cached input, which carries a 95 percent discount on the input rate, moves from $0.10 to $0.20 per million tokens on Standard. There is no tier that stays at the old price, and no published way to lock the introductory rate past November 30.

That matters because the introductory price was never meant to last. Google framed it as an introduction from the start, and the December 1 date was printed on the pricing page from launch day. Anyone who built on Argon in October and November has had about seven weeks at half price. December is when the real bill arrives.

Why the per-task math is worse than the per-token math

Per-token prices hide the real cost, because models use different amounts of tokens to finish the same job. Artificial Analysis ran Argon and Claude Opus 5.5 side by side at high effort and published the per-task result: Argon cost $1.99 per task against $1.82 for Opus 5.5. That is the figure at the introductory price, when Argon costs half as much per token as Opus. Argon already loses on cost per task while it is the cheaper model on paper.

Run that arithmetic forward to December. If the token counts hold, doubling Argon's rates takes its per-task figure to roughly $4, against $1.82 for Opus 5.5. That is our calculation on the published per-task figures, not a published number, and it assumes your tasks behave like the benchmark's tasks. But the direction is clear: the model that is already more expensive per task at half price becomes far more expensive per task at full price.

Part of the reason is thinking tokens. Argon's output price includes thinking tokens, so the bill covers the model's reasoning as well as the answer you see. A reasoning-heavy workload pays the $10 output rate, soon $20, for tokens that never reach the user. That is standard practice across frontier models, but it means the output column is where the December jump bites hardest.

What does the doubling cost in real money?

Take a small team sending 2 million input tokens and 500,000 output tokens a day through the Standard tier. In November that costs 2 times $2 plus 0.5 times $10, which is $4 plus $5, or $9 a day, about $270 a month. In December the same traffic costs 2 times $4 plus 0.5 times $20, which is $8 plus $10, or $18 a day, about $540 a month. Nothing about the workload changed. The price did.

Scale that up and the jump is a budget line, not a rounding error. A team spending $1,000 a month on Argon in November spends $2,000 in December. A team spending $5,000 spends $10,000. If your Argon bill is already a line your finance team watches, December is the month it doubles, and the time to plan for it is now, not on December 1.

What about Batch, Flex, and Priority?

Batch and Flex are the one clean escape hatch Google leaves open. Both run at half the Standard price, before and after December 1. From December, Batch and Flex cost $2 input and $10 output per million tokens, which is exactly today's Standard price. If your work can wait for batch processing or can run on flexible timing, moving it to Batch or Flex keeps your December cost at your November Standard cost.

The catch is that not all work fits. Batch is for work that can wait, and Flex is for work that tolerates flexible scheduling. A live chatbot, a real-time coding assistant, or anything a user is waiting on stays on Standard or Priority, and pays the doubled rate. Audit your calls before December and sort them into two piles: work that can wait, which moves to Batch or Flex, and work that cannot, which you price at the new rate.

Priority, the tier for guaranteed throughput, doubles too, from $3.60 and $18 to $7.20 and $36. If you are on Priority today, your December bill doubles just like everyone else's, and the premium over Standard stays the same in percentage terms.

Is there a free tier, or a way to keep the old price?

No. Argon is not available on Google's free tier, and building with it requires the paid tier of the Gemini API. There is no free allowance to fall back on, and no published grandfathering for early users. The introductory price ends for everyone on the same date.

One practical detail to watch: Google's pricing page does not state a time zone for the cutover, so treat the boundary day as uncertain and do not schedule large jobs for the last hours of November 30. A job that starts on the old rate and finishes on the new one is not a risk worth taking for a few dollars of savings.

The same December step applies if you reach Argon through GitHub Copilot, where it went generally available on October 13 and is billed at provider list pricing under usage-based billing. The price is Google's price wherever you call the model from.

How does Argon compare after December 1?

After December 1, Argon costs exactly the same as Claude Opus 5.5 on every column: $4 input, $20 output, and $0.20 cached input. The price card matches to the cent. That is a useful anchor, because it means the choice between Argon and Opus 5.5 stops being a price choice and becomes a quality choice. Pick the one that does your work better, because the bill is the same.

Against cheaper models, Argon loses on price from December. OpenAI's pricing page lists GPT-6.1 Sol at $2 input and $10 output per million tokens, a price one tracker reports as a 50 percent cut from GPT-5.6 promotional pricing on November 2. Claude Sonnet 5.5 sits at the same $2 and $10. From December 1, Argon costs twice as much as either of them on every column. If your workload runs fine on Sol or Sonnet 5.5, staying on Argon after December means paying double for the same job.

Do not confuse Argon with Gemini 4 Carbon, Google's other new Gemini 4 model. Carbon is a security-focused model announced October 28 that builds on Argon for long-horizon engineering and cybersecurity work. It carries the same introductory $2 and $10 rates, but it is not publicly available yet: Google is rolling it out first to a subset of cyber defenders, then to trusted partners, with no general public date. Carbon's price is not your price. Argon's December 1 doubling is the one that hits your bill.

What should you do before November 30?

First, measure your real usage. Pull your November token counts by tier, split input and output, and work out your December bill at the doubled rates. The arithmetic takes ten minutes and it is the only number that matters. A team that spends $50 a month on Argon can ignore the doubling. A team that spends $5,000 cannot.

Second, move what can wait. Anything that does not need an immediate answer goes to Batch or Flex before December, and stays there. That one change holds your cost at today's Standard rate for that work.

Third, use caching. Cached input costs $0.10 per million tokens now and $0.20 from December, still 95 percent below the input rate. If your prompts repeat the same context, system instructions, or documents, caching is the cheapest token you will ever buy. Note that Argon's caching is implicit only, so you cannot force a cache write. Design prompts with stable prefixes and let the cache form on its own.

Fourth, benchmark the alternatives on your own work before December. Run the same jobs on GPT-6.1 Sol and Claude Sonnet 5.5, measure the tokens each one uses and the quality of the output, and compare the per-task cost, not the per-token price. The per-token price says Argon is twice as expensive from December. The per-task cost is the number that decides whether switching saves money, and only your workload can produce it.

Who should stay, and who should switch?

Stay on Argon if your testing shows it does your specific work better than the alternatives, and the quality gap is worth double the price. A model that finishes a migration correctly on the first try is cheaper than a model that needs three retries, even at twice the per-token rate. Stay if you are on Batch or Flex and your December cost already matches today's Standard rate. Stay if your Argon bill is small enough that doubling it changes nothing.

Switch if your workload runs just as well on GPT-6.1 Sol or Claude Sonnet 5.5, because from December you would be paying twice the price for the same result. Switch if your work is batchable and you have not moved it yet, because Batch and Flex are the cheapest Argon you will ever buy. And switch, or at least re-budget, if your December Argon bill at doubled rates is a number your team cannot swallow. The deadline is November 30. The doubling is printed on Google's own pricing page. There is no surprise coming, only a bill.

Sources: Google, Gemini API pricing page, rates through November 30, 2026 and from December 1, 2026; Google, Gemini 4 Argon launch post, October 13, 2026; Artificial Analysis, per-task cost figures for Gemini 4 Argon and Claude Opus 5.5, as reported November 2026; OpenAI pricing page for GPT-6.1 Sol, as reported November 5, 2026; Anthropic pricing page for Claude Opus 5.5, as reported November 3, 2026.

Related reading: AI Basics & Tutorials, Gemini 4 Carbon Scores 88.2% on Finding Security Flaws, GPT-6.1 Terra Costs $1 Per Million Input Tokens

Frequently Asked Questions

When does the Gemini 4 Argon price increase take effect?

December 1, 2026. The introductory rates of $2 per million input tokens and $10 per million output tokens run through November 30, 2026, and the standard rates of $4 and $20 apply from December 1.

What are the new Gemini 4 Argon prices?

On the Standard tier, $4 input and $20 output per million tokens, with cached input at $0.20. Batch and Flex move to $2 and $10, and Priority moves to $7.20 and $36.

Can I lock in the old $2 and $10 prices?

No. Google has published no lock-in, grandfathering, or extension. Batch and Flex after December 1 cost the same as today's Standard rate, which is the closest thing to keeping the old price.

Is Gemini 4 Argon available on the free tier?

No. Argon requires the paid tier of the Gemini API. There is no free tier access and no free allowance for this model.

Does the output price include thinking tokens?

Yes. Google's pricing page states the output price includes thinking tokens, so reasoning-heavy work pays the output rate for tokens you never see in the answer.

What is the cheapest alternative after December 1?

On list price, GPT-6.1 Sol at $2 input and $10 output, and Claude Sonnet 5.5 at the same rates. But per-task cost depends on how many tokens each model uses on your work, so benchmark on your own jobs before switching.