Model Overview

Google introduced Gemini 3.1 Pro as the next iteration of its Gemini series, rolling it out today in a preview format for developers and consumers. The company describes the model as delivering stronger problem‑solving and reasoning abilities compared with its predecessor, Gemini 3.

Benchmark Performance

In the Humanity’s Last Exam, which measures advanced domain‑specific knowledge, Gemini 3.1 Pro achieved a record score of 44.4 percent, surpassing Gemini 3’s 37.5 percent and outperforming OpenAI’s GPT 5.2 at 34.5 percent. On the ARC‑AGI‑2 test, designed to assess novel logic challenges that cannot be directly trained, Gemini 3.1 Pro more than doubled Google’s prior score, reaching 77.1 percent compared with Gemini 3’s 31.1 percent.

Competitive Landscape

Despite the gains, Gemini 3.1 Pro does not lead the public Arena leaderboard, which reflects user preference votes on model outputs. In the text category, Claude Opus 4.6 leads by four points, while for code tasks, Opus 4.6, Opus 4.5, and GPT 5.2 High maintain a modest edge over Gemini 3.1 Pro.

Deep Think Integration

The new model also powers the latest upgrades to Google’s Deep Think tool, indicating that Gemini 3.1 Pro serves as the underlying “core intelligence” for that feature.

Implications

Google’s announcement highlights a continued focus on refining large language models for higher‑order reasoning, even as competitive benchmarks show mixed results. The preview rollout allows developers early access to test the model’s capabilities in real‑world applications, while the company continues to gather feedback for broader deployment.

Used: News Factory APP - news discovery and automation - ChatGPT for Business