Google has introduced Gemini 3.1 Pro in preview as of February 19, 2026, marking a significant upgrade in its flagship AI model series. The new iteration achieves an ARC‑AGI‑2 score of 77.1 percent—more than double that of its predecessor, Gemini 3 Pro—and records an 80.6 percent score on SWE‑Bench Verified, a benchmark focused on autonomous software engineering tasks. These results underscore a major leap in core reasoning capabilities.(en.wikipedia.org)
Gemini 3.1 Pro builds on the foundation laid by Gemini 3 Pro and the earlier “Deep Think” mode, which Google began rolling out to AI Ultra subscribers in late 2025. The new model’s performance suggests a strategic push toward AI systems capable of handling intricate, domain-specific challenges in science, research, and engineering.(en.wikipedia.org)
The ARC‑AGI‑2 benchmark measures general intelligence across a range of tasks, while SWE‑Bench Verified evaluates autonomous software engineering proficiency. Gemini 3.1 Pro’s strong performance in both areas signals Google’s intent to deliver AI models that not only reason effectively but also execute complex, multi-step technical workflows.(en.wikipedia.org)
This release positions Google ahead in the AI arms race, particularly in reasoning and task execution capabilities. As enterprises and researchers increasingly demand AI systems that can autonomously navigate complex problem spaces, Gemini 3.1 Pro may become a pivotal tool in scientific discovery, engineering design, and advanced research applications.
Looking ahead, the preview release invites developers and organizations to begin exploring Gemini 3.1 Pro’s capabilities. Its benchmark performance suggests it could soon become a core component in AI-driven innovation across technical domains.