On Thursday, Google unveiled the Gemini 3.1 Pro model, enhancing its problem-solving capabilities compared to previous versions. Despite being in preview, it excels in reasoning tasks, achieving 77.1% on the challenging ARC-AGI-2 benchmark, significantly up from 31.1% in its predecessor. While Google’s previous Gemini 3 Deep Think model surpasses it in reasoning, it’s accessible only to Google AI Ultra subscribers. Competitors like Anthropic and OpenAI have also launched recent models, with Anthropic’s Opus 4.6 scoring 68.8% and OpenAI’s GPT-5.2 at 52.9%. Gemini 3.1 Pro features a 1-million-token context window, although its output is capped at 64,000 tokens. Pricing remains competitive at $2/$12 per million tokens, making it more affordable than competitors. It’s readily available for developers via the Gemini API in Google AI Studio and other platforms. Gemini 3.1 Pro aims to lead benchmarks in reasoning and coding, establishing itself as a strong contender in the AI landscape.
Source link
