Gemini 3.1 Pro vs Gemini Flash: the price gap depends on which Flash

Specifications are near-identical; price is not

gemini-3.1-pro-low and the four Flash ids all take a 1,000,000-token context, all list text, tools and thinking, and all answer on POST /v1/chat/completions. Output ceilings are effectively the same: 65,535 tokens on Pro against 65,536 on each Flash id. Nothing in the capability column separates them.

Pro carries an official rate of $2 per million input and $12 per million output, billed at 0.1x, which is $0.20 and $1.20 per million. That is the whole of what the catalog says makes it different from the Flash line.

The Flash ids do not share one price, which is why "Pro vs Flash" cannot be answered as one question. gemini-3-flash is $0.050/$0.30 per million; gemini-3.7-flash-high and gemini-3.6-flash-high are $0.075/$0.375; gemini-3.8-flash is $0.15/$0.75. Pro is 4x the first, 2.67x/3.2x the middle pair, and 1.33x/1.6x the last.

Run the arithmetic on a job you actually have

Take a job that reads 200,000 tokens and writes 20,000 — a large-diff review, or a document pass. On gemini-3-flash that is $0.016. On gemini-3.7-flash-high it is $0.0225. On gemini-3.8-flash it is $0.045. On gemini-3.1-pro-low it is $0.064.

Read those four numbers as a spread rather than a ranking. Between gemini-3.8-flash and Pro the difference is under two cents on this job, which for most teams is inside the noise of a single retry. Between gemini-3-flash and Pro it is four times, and at a thousand jobs a day it is the difference between $16 and $64.

So the volume decides how much the decision is worth thinking about. At low volume the whole Gemini line costs about the same and you may as well pick on output quality; at high volume the same choice is a line item, and the 4x gap between the ends of the range is what you are actually deciding.

for MODEL in gemini-3.1-pro-low gemini-3.8-flash gemini-3-flash; do
  echo "== $MODEL"
  curl -s https://token-share.app/v1/chat/completions \
    -H "Authorization: Bearer $TOKEN_SHARE_KEY" \
    -H "content-type: application/json" \
    -d "{
      \"model\": \"$MODEL\",
      \"max_tokens\": 1024,
      \"messages\": [
        {\"role\": \"user\", \"content\": \"Find the race condition in this handler and explain it.\"}
      ]
    }" | jq '{answer: .choices[0].message.content, usage}'
done

What the catalog does not tell you, and how to find out

The catalog names Pro as the frontier tier and Flash as the fast tier, but it publishes no benchmark, no accuracy figure and no capability either one has and the other lacks. Every difference you can verify from it is a price. So "is Pro better on real work" is not a question this data answers, and neither is "is Flash good enough".

Both of those are answerable on your own workload, though, and cheaply. Take twenty tasks where you already know what a correct answer looks like, run them through both ids, and score the results. That gives you an accuracy difference to weigh against a price ratio you can read off the table.

One practical note while you do it: Pro is listed at low reasoning effort, and the Flash ids list their own thinking levels. If your prompts lean on a reasoning step, hold that setting constant across the comparison, or you will be measuring the configuration rather than the model.

Common questions