The year of the point release

On 19 February 2026 Google released Gemini 3.1 Pro, an update to the generation that arrived in November, with a larger context window and stronger reasoning. Like GPT-5.1, GPT-5.2 and Claude Opus 4.6 before it, this is a refinement rather than a new generation.

Four months, four decimal points, no whole numbers. That is a pattern worth naming.

What the decimal points mean

Consolidation, not stagnation. This blog made the argument on 13 November when the first of them appeared, and the intervening months have confirmed it: improvements now arrive as increments, competitors match each other within weeks, and the gap between the best and third-best model is smaller than the gap between a good and bad prompt.

The two increments that actually mattered

Of the four, two changed something practical for a small company.

The larger context windows, twice, which is why the 10 February post was worth writing: you can now put a whole document in front of the model, and that removes a real and quiet source of wrong answers.

And adaptive reasoning — the model deciding how hard to think based on the question. This is the automatic routing from 8 August, now inside the model rather than in front of it. Good for cost, and one more decision you cannot see.

What has not changed since November

The three suppliers remain within touching distance. Price and contract still separate them more than capability does. And the European data-centre answer is still the axis on which they genuinely differ.

If you ran the twenty examples in January as recommended, nothing this month requires you to run them again. Keep the file, wait for a whole number.

When to look again

The November pattern suggests providers cluster major releases before the year-end buying season, with a second cluster in spring. If a new generation is coming, it is likelier in the next two months than in the summer.

That is a forecast, not a fact, and this blog will say so if it turns out wrong.