Gemini 3.1 Pro: A smarter model for your most complex tasks
Google DeepMind
Google just rolled out Gemini 3.1 Pro, a smarter version of its main AI model, to consumers and developers. Its logic-puzzle score more than doubled versus Gemini 3 Pro, which is a big jump for a "minor" update.
Based on reporting by Google DeepMind — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Google isn't wasting any time iterating on Gemini 3. Barely a week after pushing a major upgrade to Gemini 3 Deep Think, the company is now rolling out Gemini 3.1 Pro, the model that actually powers those Deep Think gains under the hood. It's landing everywhere at once: Google AI Studio, Gemini CLI, the Antigravity coding platform, Android Studio, Vertex AI, Gemini Enterprise, and the consumer Gemini app and NotebookLM. That's a lot of surface area for a model still labeled "preview."
The number Google wants you to notice is 77.1%, its verified score on ARC-AGI-2, a benchmark built specifically to test whether a model can solve logic patterns it has never seen before rather than pattern-match against training data. Google says that's more than double what the original Gemini 3 Pro managed. Whether or not you trust benchmark scores as a proxy for real intelligence, doubling a reasoning score in a single point release is not a trivial engineering feat, and it suggests Google found real headroom in the Gemini 3 architecture rather than just tuning around the edges.
Google is pitching 3.1 Pro less as a chatbot upgrade and more as a tool for the messy middle of problem-solving: turning scattered data into one coherent view, generating visual explanations for dense topics, or carrying a creative project further than a one-shot answer would. That framing matters because it's a tacit admission that most everyday AI use doesn't need a genius-level reasoner, it needs a competent one that doesn't fall apart on multi-step tasks. Positioning 3.1 Pro as the
My take — AI-written commentary, not fact-checked reporting
I think Google is moving on a Gemini 3.x cadence now instead of waiting for a full version bump, and that's smart competitively but also a sign of how commoditized frontier reasoning gains have become. A doubled ARC-AGI-2 score sounds impressive until you remember these self-reported preview benchmarks rarely survive contact with independent testing, so I'd wait for third-party numbers before crowning this the new best reasoner.
Read more about this at: Google DeepMind