Google has introduced Gemini 3.1 Pro in preview, marking a significant advancement in its AI model capabilities. The new model, released on February 19, 2026, demonstrates a leap in core reasoning performance, achieving an ARC‑AGI‑2 score of 77.1%—more than double that of Gemini 3 Pro—and an 80.6% score on SWE‑Bench Verified, a benchmark for autonomous software engineering tasks (en.wikipedia.org).
Gemini 3.1 Pro builds on the foundation of Gemini 3 Deep Think, which Google had previously upgraded for science, research, and engineering applications (blog.google). The new model’s performance suggests a marked improvement in reasoning and problem-solving capabilities, positioning it as a more powerful tool for complex, technical tasks.
The preview release indicates Google’s continued emphasis on enhancing AI reasoning rather than focusing solely on generative fluency. By delivering substantial gains in benchmark performance, Gemini 3.1 Pro may influence how AI is applied in domains requiring rigorous logic and autonomous code generation.
While Google has not yet disclosed a general availability date, the preview signals that Gemini 3.1 Pro is entering early testing phases. Developers and researchers should monitor forthcoming announcements for broader rollout details and integration into Google’s AI ecosystem.
Overall, Gemini 3.1 Pro represents a meaningful step forward in AI reasoning, reinforcing Google’s strategy of advancing model capabilities through iterative, benchmark-driven improvements.