February 21, 2026 — Google on Thursday released a preview of Gemini 3.1 Pro, the latest iteration of its large language model (LLM), marking a significant upgrade from its predecessor, Gemini 3.

The company described Gemini 3.1 Pro as a major step forward in agentic AI capabilities and multi-step reasoning, designed to perform increasingly complex tasks across professional and real-world scenarios. The model is currently available as a preview, with general availability expected soon.

Benchmarking and Performance

Independent benchmarks have shown that Gemini 3.1 Pro significantly outperforms Gemini 3. One such benchmark, Humanity’s Last Exam, highlighted notable gains in problem-solving and knowledge application.

Brendan Foody, CEO of AI startup Mercor, which runs the benchmarking system APEX, said on social media that Gemini 3.1 Pro now sits at the top of the APEX-Agents leaderboard. Foody added that the results reflect “how quickly agents are improving at real knowledge work.”

Competitive AI Landscape

The release comes amid a heated race in the AI industry, with major players such as OpenAI and Anthropic recently unveiling new models capable of agentic reasoning and multi-step task execution. Gemini 3.1 Pro is seen as Google’s latest move to assert leadership in this rapidly evolving field.

Analysts note that Gemini 3.1 Pro could become a strong contender for enterprises seeking AI tools capable of handling complex workflows, professional problem solving, and integration into knowledge-based applications.

Google did not disclose specific technical details, but the model’s performance in early benchmarks suggests significant improvements in efficiency, reasoning, and task completion over Gemini 3.