With the next level of agentic AI requiring high-level reasoning, Google on Tuesday released Gemini 3.1 Pro to deliver it.
Google said Gemini 3.1 Pro represents the next step in core reasoning. The model is better at complex problem-solving and is meant for tasks for which a simple answer is not enough. With coding tasks, the model can generate website-ready animated scalable vector graphics (SVGs) directly from text prompts. SVGs are images created with instructions rather than pixels.
The release of Gemini 3.1 Pro follows generative AI vendor and Google rival Anthropic’s introduction on Feb. 17 of Sonnet 4.6, a model that focuses on improved coding and computer use skills. It also comes a week after Google upgraded Gemini Deep Think, its thinking model to solve challenging research problems.
The release also illuminates a trend in the AI market in which model providers are pushing toward the next step of agentic AI and touting that better reasoning will help enterprises get there. Since reasoning is a significant part of agentic AI, a model with improved reasoning skills is a superior AI system than one with subpar reasoning capabilities.
“In principle, it is true that if you want to do complex tasks, you have to be able to do better reasoning,” said Chirag Shah, a professor at the Information School at the University of Washington.
However, better reasoning is not the only building block needed for complex tasks, Shah said. Moreover, it is unclear what Google means by “complex.”
“Just saying that, ‘oh this model can do complex tasks,’ you have to take it with a grain of salt because it depends on how you define complex,” he said.
But Gemini 3.1 Pro also demonstrates other strong capabilities, such as the ability to ingest, understand, and consume data from an API, and to code a simulation using integrated tools, said William McKeon-White, an analyst at Gartner.
“This is good, continued progress,” McKeon-White said. “However, there wasn’t something I would say is a fundamental gamechanger.”
Beyond touting better reasoning skills, Google, in rolling out updated iterations of its core foundation model, Gemini, is also showing how it is building models that serve multiple enterprise needs. Although Google has yet to become known as the preferred AI vendor for a specific task, as Anthropic is known for coding, the tech giant is aiming to show enterprises that it is still well-versed in those tasks.
“They want to be the one-stop shop for all your model needs,” Shah said, adding that Google is trying to build a portfolio that will appeal to enterprises that might care about coding, searching and informational tasks. While enterprises might have the option to choose from Anthropic, OpenAI, Google or other vendors, most will settle on only one ecosystem to buy into. Google is trying to provide well-crafted models that fit different enterprise needs.
