Google DeepMind

Google's Pro line has stopped shipping and its cheap line has not. In the OpenRouter catalog on 3 September 2026, the newest Gemini row on the Pro line is google/gemini-3.1-pro-preview, listed on 19 February 2026 and still carrying preview in its name. The Flash line has shipped four times since: google/gemini-3.5-flash in May, google/gemini-3.6-flash on 21 July, google/gemini-3.7-flash on 13 August, google/gemini-3.8-flash on 2 September. llm-releases counts that run as Google's fourth Flash model in under four months; 9to5Google read it as the third Flash update in three months, with 3.8 rolling out three weeks after 3.7.

The scoreboard moved with the shipping. The Artificial Analysis intelligence index the catalog carries puts google/gemini-3.8-flash at 41.2openrouter-models, last checked 2026-09-14, against 39.4openrouter-models, last checked 2026-09-14 for the August release and 34.3openrouter-models, last checked 2026-09-14 for July's — two climbs in the six weeks between them. The Flash row lists at $0.75per million tokensopenrouter-models, last checked 2026-09-14 input; the Pro preview, six months older, lists at $2.00per million tokensopenrouter-models, last checked 2026-09-14.

The discount arrived before the capability did. google/gemini-3.5-flash from May listed at $1.50per million tokensopenrouter-models, last checked 2026-09-14; google/gemini-3.6-flash in July halved that to $0.75per million tokensopenrouter-models, last checked 2026-09-14, and the two releases since have held it there. The jump on the index came twenty-three days after the price cut, on a row that costs the same.

Google says the latest jump is a design choice: the launch post explains that "3.8 Flash works harder", executing extra reasoning steps and calling tools iteratively, and that "at times, the model might use more tokens to maximize performance, especially at higher effort levels". llm-releases prices that sentence: output tokens per task rose ~30%, to ~48ksource, accessed 2026-09-03, lifting cost per task to ~$0.58source, accessed 2026-09-03 at high reasoning even though the per-token price never moved.

The price on the current row is dated in a way vendors rarely publish: Google's launch post calls it introductory pricing, names 31 December 2026 as the day it ends, and states the regular price that follows — double the introductory one on both input and output. A screenshot of today's price sheet is wrong from 1 January 2027 onward, and says nothing about it.