Google Gemini 3.1 Pro: twice the reasoning, same price

On February 19, 2026, Google quietly shipped Gemini 3.1 Pro: there was no big launch show around it, just a silent update on the model list. The benchmark tables told a different story: according to Google's own numbers, it delivers roughly double the reasoning performance of Gemini 3 Pro, and it tops the leaderboard on 12 out of 18 tracked benchmarks. If your team runs a Gemini-based workflow, this upgrade landed in your lap without you having to switch models.

What happened

Gemini 3 Pro was already a strong reasoning model on its own, but the 3.1 release specifically targeted multi-step inference: the kind of task where the model doesn't just have to spit out an answer, but has to carry a chain of reasoning through to the end without errors. According to Google's messaging, this "double reasoning" claim isn't the result of one narrow test, it shows up consistently across most of the tracked benchmarks.

The gist in one minute

• Release date: February 19, 2026
• Headline claim: roughly 2x reasoning performance compared to Gemini 3 Pro
• Benchmark position: #1 on 12 out of 18 tracked tests
• Pricing and availability didn't change in any meaningful way according to the announcement; it lands in the same place Gemini 3 Pro already lived

What it can do: the benchmarks

Google didn't break down the actual benchmark numbers in as much depth as a full technical report usually would, and that alone is telling: the big AI labs are increasingly shipping "quiet" updates where the marketing is minimal but the benchmark results still end up leading the market. This matters for teams that pick models programmatically through an API: it's worth re-measuring periodically, because the "best model" spot can shift under you week to week without you doing anything.

"Google quietly doubled it, and we quietly rewrote a few config files. This is exactly the right amount of excitement for me, nothing dramatic, just one more line in the changelog." (Nasus)

Who should care

Mainly teams already working inside the Gemini ecosystem: Google Cloud integration, Vertex AI, or simply having the Gemini API already wired into a pipeline). If your team runs reasoning-heavy tasks (code review, complex document analysis, multi-step agentic workflows), this update alone is reason enough to re-run your own benchmarks comparing the old and new model. If you live in a different ecosystem, treat this as a signal instead: the race isn't slowing down, and "which model should we use" is a question worth revisiting every quarter, not something you decide once and forget.

Source: https://gemini3.us/gemini-3.1-pro