On February 19, 2026, Google released Gemini 3.1 Pro in preview. The headline number is blunt: 2x reasoning versus Gemini 3 Pro, at the same pricing. The API endpoint name is equally blunt: gemini-3.1-pro-preview.
Doubling Capability at the Same Price
“Twice the reasoning, same price” is a classic price-performance statement. As the capability gaps between frontier models get harder to feel, Google is speaking in capability-per-dollar — for teams that call APIs at volume, a more practical comparison axis than any single benchmark.
It is also an old cloud-era playbook: push price-performance to a position rivals struggle to follow, and absorb the cost difference with scale. The model market is inheriting that strategy, and Google is among the best-positioned players to run it — infrastructure scale is its home field.
For buyers, the arithmetic lands directly in unit economics. If reasoning quality per dollar doubles at the same list price, workloads that were borderline uneconomic — long chains of thought, high-volume evaluation, exhaustive code review — cross into the affordable column without any negotiation. Price cuts get headlines; capability-at-price changes what teams choose to build.
What the Preview Endpoint Means for Builders
The -preview suffix in the endpoint name is an explicit signal to API developers:
- Good for evaluation and trial runs — put the new model through your own eval pipeline and quantify the effect on real workloads
- Not good for production commitments, since preview behavior and timelines can shift
- Production traffic should stay pinned to stable endpoints, with migration managed as a proper project
Version lifecycle management is now a basic skill of building on LLMs. Google drawing the line through naming alone is one of the friendliest ways to do it.
A disciplined pattern for preview adoption: run the preview endpoint in shadow against your current model on real traffic samples, score both with your own rubric rather than public benchmarks, and time the production switch to a stable release. Preview access is essentially free information — the only cost is the eval harness you should already have.
Where This Lands in February’s Cadence
Place Gemini 3.1 Pro back into the month’s release sequence: Claude Opus 4.6 and GPT-5.3-Codex arrived on the same day, February 5; GLM-5 went open source on February 11; Claude Sonnet 4.6 took the default slot on February 17. Four major camps all played their cards within two weeks. The pace also compresses evaluation cycles: a team that takes three weeks to benchmark a new release now ships its verdict after the next release has already landed.
In a cadence that homogenized, Google’s differentiating weapon is the clearest: pricing backed by infrastructure scale. Double-at-the-same-price is not a short-term promotion — it moves the competitive axis from “who is strongest” to “who is strongest per dollar.” That fight is just starting. It also raises the bar for everyone else’s pricing pages: once the market’s reference point is doubling at flat cost, a rival shipping a modest improvement with a price increase has to justify itself twice — on capability and on cost.
Sources
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.