Guide

Choosing a Gemini model: five ids on one price ladder

Every Gemini id here takes 1,000,000 tokens, lists the same capabilities and uses the same route. Prices run from $0.050 to $0.20 per million input — a 4x spread with nothing else changing.

The specifications are the same across all five

gemini-3.8-flash, gemini-3.7-flash-high, gemini-3.6-flash-high, gemini-3-flash and gemini-3.1-pro-low all take a 1,000,000-token context. All five list text, tools and thinking. All five answer on POST /v1/chat/completions. Output ceilings are 65,536 tokens on the four Flash ids and 65,535 on the Pro id.

That is unusual for a model family and it simplifies the decision considerably. There is no id here that fits a prompt the others cannot, no id that supports tool calling while another does not, and no route change to write code around.

What is left is price, and the spread is wide. Per million input tokens after each multiplier: $0.050 on gemini-3-flash, $0.075 on gemini-3.6-flash-high and gemini-3.7-flash-high, $0.15 on gemini-3.8-flash, $0.20 on gemini-3.1-pro-low. Output runs $0.30, $0.375, $0.75 and $1.20 in the same order.

Read the ladder, not the version numbers

The ordering by price is not the ordering by version, and this is where most of the confusion on this family comes from. gemini-3.8-flash is the newest id (2026-09-02) and the second most expensive, above gemini-3.7-flash-high and gemini-3.6-flash-high, which are both older and both half the price.

The reason is the multiplier rather than the rate card. gemini-3.8-flash, gemini-3.7-flash-high and gemini-3.6-flash-high all carry the same $0.75/$3.75 official rate; the first is billed at 0.2x and the other two at 0.1x. Identical upstream pricing, doubled bill on one of the three.

gemini-3-flash sits below all of them on a genuinely lower official rate — $0.50/$3, at 0.1x — which puts it at two thirds of the 3.6/3.7 pair on input and four fifths on output. And gemini-3.1-pro-low tops the ladder at $2/$12 official, also at 0.1x.

One loop covers the whole family; only the model field changes.
for MODEL in gemini-3-flash gemini-3.7-flash-high gemini-3.8-flash gemini-3.1-pro-low; 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\": 512,
      \"messages\": [
        {\"role\": \"user\", \"content\": \"Explain what this SQL query does and name one risk.\"}
      ]
    }" | jq '{answer: .choices[0].message.content, usage}'
done

How to pick without a benchmark to lean on

The catalog gives you prices, dates and specifications. It does not give you accuracy scores, and we are not going to substitute a guess for one. So the decision has to be made on your own output, and the identical request shape across all five ids makes that a short experiment rather than a project.

A workable procedure: assemble twenty tasks from your real workload where you know what a good answer looks like. Run them through gemini-3-flash first, since it is the bottom of the ladder, and count the failures. If there are none you can see, everything above it is spend without evidence on your workload. If there are, move up one rung and repeat.

Keep the reasoning configuration fixed while you do it — the Flash ids list their own thinking levels, and gemini-3.1-pro-low is the low-effort Pro id — otherwise you are comparing settings. And check the token totals from that run against the ladder: at low volume the whole family costs about the same, and at high volume the 4x between the ends is the entire decision.

Common questions

Why is the newest Gemini id not the most expensive one?
Because price here is the official rate times each model's own multiplier. gemini-3.8-flash is 0.2x on a $0.75/$3.75 rate, giving $0.15/$0.75 per million. gemini-3.1-pro-low is 0.1x on a $2/$12 rate, giving $0.20/$1.20 — higher, despite the lower multiplier and the older release date of 2026-03-01.
Is there any reason to pick gemini-3.6-flash-high over 3.7?
Nothing in the catalog distinguishes them on price or specification: both are $0.075 per million input and $0.375 output, both take 1,000,000 tokens with a 65,536-token ceiling, both list text, tools and thinking. The only recorded difference is the release date, 2026-07-21 against 2026-08-13. If you have an existing integration pinned to 3.6 and it works, the table gives you no cost argument for moving.
Do all five Gemini ids really have a 1M context window?
Yes — 1,000,000 tokens on every one, including gemini-3-flash at the bottom of the price ladder. Output ceilings are 65,536 tokens on the four Flash ids and 65,535 on gemini-3.1-pro-low. Paying more within this family does not buy more room.