Why Google is still on Gemini 3.1 Pro
Why Google is still on Gemini 3.1 Pro
The workhorse is on 3.8. The model with Pro in the name has not moved since winter.
Open Google's API catalog and the Pro row is still Gemini 3.1 Pro Preview. Not 3.5. Not 3.8. The same checkpoint that has been the documented Pro model since late February, priced at $2 / $12 per million tokens up to 200k and $4 / $18 above that.
Flash did not sit still. From the May 19 I/O launch of 3.5 Flash through 3.8 Flash on September 2, Google shipped four numbered workhorses. That cadence is its own piece. This one is only the Pro question: why the flagship name is still 3.1.
3.5 Pro was supposed to replace it in June
At I/O, Google said 3.5 Pro was already in internal use and would roll out the next month. Sundar Pichai told developers June. On July 21, in the 3.6 Flash post, the line had changed: 3.5 Pro was in partner testing and would ship when it was ready. The same paragraph announced that Gemini 4 pretraining had started.
As of this month, DeepMind's public pages still call 3.5 Pro "coming soon." There is no gemini-3.5-pro model id in the catalog developers can call. The Pro id you can call is 3.1.
That is a miss of a named month, then a summer of "when it's ready," while Flash versions kept landing on a schedule.
They will not ship a Pro that loses to Flash
3.5 Flash's own launch said it beat Gemini 3.1 Pro on the benchmarks Google wanted quoted: Terminal-Bench 2.1 at 76.2%, plus GDPval-AA and MCP Atlas, at about four times the output speed of other frontier models. It became the default in the Gemini app, AI Mode in Search, Antigravity, and Spark.
Then Flash kept climbing. On DeepSWE v1.1, the 3.8 model card shows 73.7%, next to Claude Opus 5 at 74.0%. A Pro launch in September that cannot clear the Flash already running in Search is a press cycle Google has already lived through once. Reporting after the June miss, including Decrypt's account of Bloomberg, said the larger model fell short internally, especially on coding, and a late-June training-data refresh disappointed.
So 3.1 Pro stays. It is the last Pro checkpoint Google was willing to put a public name on. Replacing it with a 3.5 that loses the coding comparison to 3.8 Flash would make the word Pro mean "older and more expensive." Leaving 3.1 in the catalog at least means Pro means "the stable larger model," even if Flash is what actually moved.
3.1 Pro and Gemini 4 are different delays
It is easy to treat "no new Pro" as "Gemini 4 is late." Google has described two tracks, and they are not the same model.
3.5 Pro is a Gemini 3 flagship that missed June and is still described as partner testing. Gemini 4 is the pretraining run announced on July 21 as the most ambitious they have started. No size, no price, no API id, no quarter. Google has not said 4 replaces 3.5 Pro, and it has not said 3.5 Pro is cancelled.
3.1 Pro is what you get while both of those are unfinished. It is a preview that became the production Pro by default, because nothing after it cleared the bar to take the name.
Flash could move every three weeks because 3.7 is based on 3.6 and 3.8 is based on 3.7. A new Pro is a new base, or it is a 3.5 they already decided was not ready. Neither of those ships on a Flash calendar.
What to call in a review
Use 3.1 Pro when you want that specific checkpoint: the longer-context price tier, the model your evals were pinned to in the spring, a second opinion that is not the Flash default. It is still a real model. It is not the model Google has been improving in public since May.
Use 3.8 Flash when you want the Gemini that actually absorbed the summer. Same introductory price as 3.7, $0.75 / $3.75 per million tokens through December 31, and the coding scores Google is willing to print next to Opus. It is in ThinkReview Model selection.
Do not wait on 3.5 Pro for a merge request that is open now. June was the date. September still does not have the id. If a Pro-branded successor shows up, the test is whether it beats 3.8 Flash on the diff in front of you, not whether the name sounds newer than 3.1.
ThinkReview runs that comparison on GitHub, GitLab, Azure DevOps, and Bitbucket. Pick 3.8 when you want the workhorse Google kept shipping. Pick 3.1 Pro when you mean the flagship they refused to replace.
The June promise and the July update are in Google's 3.5 and 3.6 posts. DeepSWE figures are from the 3.8 model card. Review with the Gemini that shipped: ThinkReview.