Three vendors now list their newest model at $2 per million input tokens and $10 per million output. Only one of them has said the price will double. Gemini 4 Argon pricing looks like a tie on launch day, and we think the tie is temporary. Budget at the later rate, and cap output length before anyone on your team touches it.
Google’s launch post puts Argon at the same sticker as OpenAI’s GPT-6.1 Sol and Anthropic’s Sonnet 5.5, both at $2 and $10 today. A matching price on launch day tells you very little about the bill in January.
Key points
- Is Gemini 4 Argon pricing really a three-way tie? Only for now. Argon lists at $2 per million input tokens and $10 per million output, and Google says the rate rises to $4 and $20 after the introductory period.
- What does one maximum-length run cost? At the introductory rate, about $11. After it, about $22.
- What does the doubling do to an ordinary month? Take a small team that sends 5 million input tokens and gets 2 million output tokens back in a month.
- Whose benchmarks are these? Google’s. It reports 77.9% on DeepSWE v1.1, 91.7% on the LVBench long-video test, 51.3% on Zapier’s AutomationBench for the top spot, and 68% on CWE-bench v1 for vulnerability fixes, tied for first.
Is Gemini 4 Argon pricing really a three-way tie?
Only for now. Argon lists at $2 per million input tokens and $10 per million output, and Google says the rate rises to $4 and $20 after the introductory period. Cached input is 95% cheaper. We haven’t seen an end date, so any plan you make today is a plan for a quote that isn’t final.
We made the same point about Claude Opus 5.5’s price cut, where the effort setting mattered more than the list rate. Build a workflow on the introductory rate and you’ve built it on a number Google already told you will double.
Access is the other catch. Google says the model goes first to trusted cyber defenders through its Fairwind Program, with paid API customers and Google AI Ultra subscribers next, after more testing. Most small firms can’t buy it yet.
What does one maximum-length run cost?
At the introductory rate, about $11, and about $22 after it. That is our arithmetic, not a measurement. Say a job reads 500,000 tokens and writes 1,000,000. Input costs $1.00 now and $2.00 later, while output costs $10.00 now and $20.00 later. Output is where the bill sits on long runs.
Earlier Gemini generations stopped at 64,000 tokens per response. Argon can write 1 million in one run, which suits rewriting a codebase rather than summarising one. Cap the same job at the old ceiling and the output costs about $0.64 now, or $1.28 later. Ten full-length runs a week is $110 now and $220 later.
That’s fine for a software team replacing a legacy module. It’s an ugly surprise for an office that switched on long outputs because they were available. Output tokens are the expensive ones, and a bigger ceiling means a longer answer unless somebody sets a limit. Set it in the settings before you hand the keys to anyone.
What does the doubling do to an ordinary month?
Take a small team that sends 5 million input tokens and gets 2 million output tokens back in a month. The figures are ours, picked to be round. At the introductory rate that’s $10 for input and $20 for output, $30 in all. At the later rate it’s $20 and $40, $60 in all. Small money either way.
Per-token prices mislead in the other direction too, as GPT-5.5 pricing shows when you price by the finished task.
Now scale the habit, not the price. If long outputs become routine and the same team writes 20 million output tokens a month, the bill is $200 now and $400 later on output alone. The doubling is a fixed multiplier. Output volume is the variable, and it’s the one your staff control.
That’s why the cap matters more than the rate card. Nobody on your team negotiates Google’s price. Everybody on it can accept the default length.
Whose benchmarks are these?
Google’s. It reports 77.9% on DeepSWE v1.1, 91.7% on the LVBench long-video test, 51.3% on Zapier’s AutomationBench for the top spot, and 68% on CWE-bench v1 for vulnerability fixes, tied for first. These are the vendor’s figures, picked by the vendor.
The choice of tests tells you something. Google led with security, automation and long-running agent work, where public evaluations are thinnest and a company’s own judgment counts most. That doesn’t make the results wrong. It means nobody outside Google has checked them yet, and we wouldn’t buy on them.
A Bloomberg report relayed by Android Headlines adds a caution. Some Google employees with model access reportedly find Argon strong on benchmarks and weaker on messy, real-world coding and front-end design. Google denies any gap. A launch table can’t settle that, and neither can we from here.
For a buyer the useful reading is plain. Treat the vendor’s table as a list of what Google is proudest of, not a ranking of what you’ll get. Your tasks are probably duller than a vulnerability benchmark, and dull tasks are where models differ in ways a table won’t show.
What should a buyer do before committing?
Price one real task end to end on each vendor and keep the invoice. Use real work, not a demo prompt. A job your team does every week, run through all three at their current prices, then costed again at Argon’s later rate. Measure cost per successful task, not cost per token.
The same lens applies to the model-specific claims we checked in our look at the Jev model. For how the three vendors compare on features rather than cost, our ChatGPT, Claude and Gemini business comparison is the better start. Pricing is half the test, and a narrow model like the Jev AI model shows the other half is what it refuses to do. Launch prices are one trap, and vendor-run benchmarks like Kimi K3‘s are another, so test on your own tasks.
A fair objection is that a 1 million token output is a spec for labs, not for a 40-person firm, and that comparing prices before the model is on sale is a distraction. Both points land. Most readers won’t generate anything near a million tokens. But a default setting or an eager employee can get you there without anyone deciding to, and that’s the case the cap is for.
We’d be wrong about the bill if Google’s final pricing stayed at $2 and $10, or if long output turned out to be metered in a way that kept typical jobs cheap. We’d be wrong about the model if independent tests on ordinary business code showed the benchmark wins carry over. None of that has happened, and the model wasn’t generally available on the day we looked. Prices have changed within weeks across this industry, so read every figure as correct on October 8, 2026, and likely to move.
Three vendors, one sticker price, one footnote. Read the footnote.
Frequently asked questions
How much does Gemini 4 Argon cost?
Google lists $2 per million input tokens and $10 per million output tokens at launch, rising to $4 and $20 after the introductory period. Cached input is priced 95% lower.
Who can use Gemini 4 Argon today?
Google says trusted cyber defenders get it first through its Fairwind Program. Paid API customers and Google AI Ultra subscribers follow after more testing.
Is the 1 million token output worth paying for?
Only for jobs that need very long answers, like large code migrations. For ordinary tasks, cap the output length and compare cost per finished task rather than per token.
Archive entry dated October 1, 2026, written and fact-checked on October 8, 2026. Sources are linked on the claims they support.
