Gemini 3.8 Flash vs 3.7 Flash: same rate, twice the multiplier

Same rate card, different multiplier

gemini-3.8-flash and gemini-3.7-flash-high have the same official rates: $0.75 per million input tokens and $3.75 per million output. They also share their specifications — a 1,000,000-token context, a 65,536-token output ceiling, text, tools and thinking support, and the POST /v1/chat/completions route.

The multipliers differ. gemini-3.8-flash is billed at 0.2x, gemini-3.7-flash-high at 0.1x. Multiply out and the newer id costs $0.15 per million input and $0.75 per million output; the older one costs $0.075 and $0.375. Exactly double, on both directions of the bill.

gemini-3.6-flash-high sits alongside 3.7 on the same terms: the same $0.75/$3.75 official rate, the same 0.1x multiplier, the same 1,000,000-token window and 65,536-token ceiling. So the catalog holds two ids at $0.075/$0.375 and one at $0.15/$0.75, and the expensive one is the recent one.

What doubling looks like on a real month

Percentages are easy to wave away, so put tokens on them. A month of 5,000,000 input tokens and 1,000,000 output tokens costs $0.75 on gemini-3.7-flash-high and $1.50 on gemini-3.8-flash. A single large job of 1,000,000 input and 200,000 output is $0.15 against $0.30.

The ratio holds at every volume because both sides of the rate card double together — there is no mix of traffic that makes the gap narrower or wider. Whatever your input-to-output ratio is, 3.8 Flash is 2x the bill of 3.7 Flash.

That is unusually clean as trade-offs go. You are not weighing a cheaper input against a dearer output; you are deciding whether the 20 days between the 2026-08-13 and 2026-09-02 release dates bought something your workload can detect.

for MODEL in gemini-3.8-flash gemini-3.7-flash-high; do
  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\": \"Review this diff for correctness bugs.\"}
      ]
    }" | jq -r '.choices[0].message.content'
done

The catalog cannot answer whether 3.8 is better

Everything the catalog records about these two ids is equal except the multiplier and the release date. It lists no benchmark, no quality figure, and no capability that one has and the other lacks. Anyone telling you the newer id reasons better is not reading this table, and we are not going to assert it on their behalf.

What that leaves is a measurement you have to take yourself, and it is cheap to take. Both ids answer on the same route with the same body, so a comparison is a loop over two strings. Run a representative slice of your own prompts through each and look at the outputs side by side.

Set the bar before you run it, not after: if 3.8 Flash does not produce a difference you can point at on your own work, you are paying 2x for a version number. If it does, the doubled rate is still only $0.15 and $0.75 per million, and the decision is yours to make against your own volume.

Common questions