Gemini 3.1 Pro
Gemini 3.1 Pro ships as gemini-3.1-pro-preview and its price is tiered: under 200,000 tokens it is $2 in and $12 out per million, and at 200,000 and above it becomes $4 and $18. Our coding table does not rank it, because eighty-five of a hundred points of weight have no verified evidence behind them.
- preview model ID
- price tier at 200k tokens
- unranked in our table
Last checked: This is a reference page. It is re-checked against the vendor sources and updated when a new version ships. What changed
Tiered pricing, with the boundary at 200,000 tokens
Most models in this section have one price. Gemini 3.1 Pro has two. Below 200,000 tokens, input runs at $2 and output at $12 per million. At 200,000 and above, input becomes $4 and output $18: double on input, half again on output.
One thing we could not settle from Google's pricing page, and will not guess at: whether crossing the boundary re-rates the whole request or only the tokens above it. The page lists two tiers and stops there. If your architecture builds long requests, that distinction is not small on an invoice and you should test it on a real call.
Why our coding table does not rank it
The table's rule is that a candidate missing more than half its criteria by weight is not ranked and appears below the table under its own name. Gemini 3.1 Pro is missing eighty-five points of a hundred, and it is the only candidate in that position.
Three reasons, and all three are fixable. We have not read its row on the WebDev Arena leaderboard, and that column alone carries forty points. Google's models page prints no context window at all, so the ten-point context column sits empty and we do not write a number we have not seen. And we have not assigned this version a tooling-maturity score, which is fifteen more. Its only evidence today is a row on the general text leaderboard at 1486.
None of that is a verdict on the model. A table that fills an empty cell with a plausible number is worse than one that shows the cell empty.
Preview, and the Iran column
The ID Google publishes ends in preview. You would be building on something its maker has not called final: the ID can change, behaviour can shift without an announcement, and the price can differ in the stable release. For prototyping that is fine. For something that has to still work in six months it is not.
This page's Iran column comes from the Gemini API available-regions list. That page names more than 190 countries and territories and Iran is not among them. Something visible across the five models we wrote this week: only Google and Anthropic publish such a list at all. Moonshot, Z.ai and DeepSeek publish none, so the Iran column for those three stays unchecked, and a vendor that documents its restrictions ends up ahead of one that documents nothing.
Using it from Iran
This column carries a date and says how it was checked, because most listicles guess it. Where we have only read a vendor policy page, the note below says exactly that.
- Reachable
- blocked
- Payment
- no working route
- Free tier
- no
- Measured on
How we checked: The Gemini API available-regions page lists the countries where the API works, and Iran is not on that list. We read it there rather than measuring it: an independent test has not been run yet.
What it is good at
- The first price tier, $2 and $12, undercuts both Opus 5 and Fable 5
- It scores 1486 on the general text leaderboard, inside the top cluster
- Google publishes an available-regions list for the API, so the access column is not a guess
Where it falls short
- Our coding table cannot rank it: eighty-five of a hundred points have no verified evidence.
- Google prints no context window for this model on its models page, so we print none either.
- We have not read its row on the web-dev leaderboard, and that column is forty points of the table.
- Its ID ends in preview, so behaviour and price can change without an announcement.
- It is unclear whether the above-200k rate applies to the whole request or only to the tokens above the boundary; the pricing page lists two tiers and no more.
- Iran is not on the Gemini API available-regions list.
Our take
While the ID says preview and the context window is unpublished, we would not put this model underneath a product. For testing and comparison, yes. If Google exposes a web-dev row and a context figure, it goes back into the table the same day and earns its own rank.
Questions people actually ask
What does Gemini 3.1 Pro cost
Under 200,000 tokens, $2 per million input tokens and $12 per million output. At 200,000 and above, $4 and $18.
What is the Gemini 3.1 Pro context window
Google does not publish that number on the Gemini API models page, so we do not write one. Any figure you have seen elsewhere did not come from a primary source.
Why is it unranked in your coding table
Because eighty-five of a hundred points have no evidence: the web-dev leaderboard row, the context window, and a tooling score. Our rule leaves a candidate unranked when more than half the criteria by weight are empty.
What does a preview ID mean
That Google has not called this version stable. The ID, the behaviour and the price can all change before the final release, which is a risk for a system that has to run for months.
Sources
- Gemini API models vendor source 12 August 2026
- Gemini API pricing vendor source 12 August 2026
- Gemini API available regions vendor source 12 August 2026