Quick answer
To estimate your Gemini 4 Argon API cost, measure the average input and output tokens for one real job, multiply by your monthly job count, then apply the reported intro rates (about $2 per million input tokens and $10 per million output tokens) and the planned rates from December (about $4 and $20). Budget at the higher figure, because the intro window is temporary.
Key takeaways
- Reported intro rates: about $2 input / $10 output per million tokens
- Planned from December: about $4 / $20, so budget for double
- Output tokens usually drive the bill
- Measure tokens on a sample of real jobs before you commit
In this guide
- What are the reported prices?
- What is the cost formula?
- How do you measure tokens per job?
- What does a worked example look like?
- How does the 1 million token output affect cost?
- How should you compare Argon with other models?
- What should you do before committing?
- Frequently asked questions
- The bottom line
This guide answers one question from our article Gemini 4 Argon Review: Has Google Finally Caught Up?: what will Argon actually cost you through the API, now and after the intro pricing ends?
📘 Main article: This guide supports our full article, Gemini 4 Argon Review: Has Google Finally Caught Up?. Read it for the complete picture.
What are the reported prices?
According to Matt Wolfe's review, as summarised in our main article:
| Period | Input (per million tokens) | Output (per million tokens) |
|---|---|---|
| Introductory, to end of November | About $2 | About $10 |
| Planned from December | About $4 | About $20 |
Those are the rates reported at the time. The planned change roughly doubles both prices, so any estimate should show both figures side by side.
⚠️ Warning: These prices are as reported in Matt Wolfe's review at the time. Check Google's current pricing page before you set a budget or sign off a project.
What is the cost formula?
The calculation is the same for any token-priced model:
- Monthly input tokens = average input tokens per job × jobs per month
- Monthly output tokens = average output tokens per job × jobs per month
- Monthly cost = (input tokens ÷ 1,000,000 × input price) + (output tokens ÷ 1,000,000 × output price)
That is all you need. The hard part is getting realistic token counts.
How do you measure tokens per job?
- Run 10 to 20 real jobs through the API, not made-up examples
- Record the input and output token counts the API reports for each
- Use the average, and note the largest job too
- Include any system instructions or documents you send each time, because they count as input
Output length varies more than input, especially for long-running work. A short summary and a long report can differ by a factor of a hundred.
What does a worked example look like?
Imagine a small business that generates product research reports. Each report uses about 20,000 input tokens and 5,000 output tokens, and you run 500 reports a month.
| Item | Calculation | Intro rates | December rates |
|---|---|---|---|
| Input | 10 million tokens | $20 | $40 |
| Output | 2.5 million tokens | $25 | $50 |
| Total per month | $45 | $90 |
These are illustrative numbers to show the method, not a quote. Your own jobs will differ.
Notice that output costs five times as much per token as input. Long outputs drive the bill.
Budget at the higher rate. If a workflow only makes sense at intro pricing, it probably does not make sense yet.
How does the 1 million token output affect cost?
Argon's standout feature is its reported 1 million token output limit, roughly 750,000 words. That is powerful for very long artefacts, but a single maximum-length output at the December rate would cost about $20 on its own.
The 1 million token ceiling is a capability, not a target. Set sensible output limits for routine jobs and save very long outputs for work that genuinely needs them.
💡 Tip: Just because Argon can write up to 1 million output tokens does not mean it should. Set a maximum output length for routine jobs to keep costs predictable.
How should you compare Argon with other models?
Price per token is only part of the picture. Our main article noted that Argon showed a lower cost per task than many peers on one chart, and that its intro API prices sat at roughly half of Opus and around Sonnet 5.5 and GPT-6.1 Sol levels.
To compare fairly:
- Run the same jobs through each model
- Compare cost per finished job, including retries and edits
- Use the settings you would really use
- Recalculate Argon at the December rate
Our guide to comparing models fairly across effort levels explains why settings matter so much.
What should you do before committing?
- Calculate both intro and December costs
- Set a monthly spending alert
- Cap output length for routine jobs
- Re-check Google's pricing page before launch
Want more grounded AI cost guides? We turn hands-on reviews into practical steps for small businesses. Browse the latest on Prodify Digital.
Frequently asked questions
How much does the Gemini 4 Argon API cost?
Reported intro rates were about $2 per million input tokens and $10 per million output tokens, with plans to rise to about $4 and $20 after November.
Why does output cost more than input?
Generating text is more expensive for the provider, so most models charge more for output tokens.
Is Argon cheaper than Opus 5.5?
At the reported intro rates it was roughly half the price of Opus per token. Compare cost per finished job on your own tasks.
Do Gemini app subscriptions include API usage?
Our main article does not say they do. Treat consumer plans and the API as separate unless Google confirms otherwise.
How often should I re-estimate?
Whenever prices change and whenever your job volume or length changes noticeably.
The bottom line
Measure real jobs, apply the simple formula and always show the December figure next to the intro price. If the numbers still work at the higher rate, Argon's API is worth a proper trial.
Related posts
- Gemini 4 Argon Review: Has Google Finally Caught Up?
- Google AI Pro vs Ultra for Gemini 4 Argon: Which Plan Do You Need?
- Claude Effort Levels Explained: How to Compare Models Fairly
[…] Thinking about switching to a cheaper model? Start with a fair side-by-side test on one real task, and see how we estimated costs in our Gemini 4 Argon API cost guide. […]