Opus 5 or Gemini 3.1 Pro: a comparison half of which is unpublished
Gemini 3.1 Pro is cheaper, $2 and $12 against $5 and $25, but only under 200,000 tokens; above that it becomes $4 and $18. Opus 5 carries one price across its whole one million token window, and Google publishes no context window at all for this Gemini model.
Specifications where both sides have a verified number
| Claude Opus 5 | Gemini 3.1 Pro | |
|---|---|---|
| API model ID | claude-opus-5 | gemini-3.1-pro-preview |
| Input price | $5 per million tokens | $2 per million tokens |
| Output price | $25 per million tokens | $12 per million tokens |
A row where one side has no verified number is left out entirely, because readers read a half-empty row in favour of the filled side.
Why the table above has three rows
Because most of the specifications you would want to compare are unpublished on one side. The Gemini API model page carries no context window, no output ceiling and no knowledge cutoff for gemini-3.1-pro-preview. Opus 5 publishes all three. A row with one empty cell does not get built, because readers read a half-empty row in favour of the filled side.
Tiered pricing, and where "half the price" breaks
The Gemini figures in the table are the tier below 200,000 tokens. At 200,000 and above, input becomes $4 and output $18. So exactly where a large window earns its keep, the price gap closes from "less than half" to "about three quarters". Opus 5 has no such step; its $5 and $25 hold to the end of the window.
One is ranked, the other is not
In our coding table Opus 5 is first with evidence in all six columns. Gemini 3.1 Pro takes no rank at all, because 85 of its 100 points of weight are empty: the web-dev leaderboard row, the context window and the tooling maturity score. Its only complete column is price. Do not read that as a judgement on the model; it is the state of the published data, and a model still in preview is entitled to be incomplete.
Pick this if Claude Opus 5
- You want to know what you are buying: window, output ceiling and knowledge cutoff are all published.
- Your work is code. Opus 5 is first in our table and Gemini 3.1 Pro is not ranked at all.
- You send long inputs and do not want the price to step up mid-job.
Pick this if Gemini 3.1 Pro
- Cost per token decides it and your prompts stay under 200,000 tokens.
- The rest of your pipeline is on Google and you want the same account.
- You are comfortable with a model still in preview whose full specification is unpublished.
Questions people actually ask
Which is cheaper
Gemini 3.1 Pro, conditionally. Under 200,000 tokens it charges $2 and $12 against $5 and $25 for Opus 5. From 200,000 tokens up it becomes $4 and $18 and the gap narrows sharply.
Why is Gemini 3.1 Pro unranked in your coding table
Because more than half the criteria weight has no evidence behind it: we have not read its web-dev leaderboard row, and Google publishes neither a context window nor anything we can score for tooling. Our rule leaves a candidate unranked when more than half of it is empty.