Gemini 3.1 Pro Preview

Gemini 3.1 Pro Preview, released February 2026, is Google's frontier reasoning model and one of the most capable multimodal AI systems available. It features a 1M token input context (1,048,576 tokens exactly), 64K max output tokens, and native multimodal support for text, images, video, audio, and PDF inputs — making it uniquely versatile for complex real-world tasks. It is available across major model platforms and developer tools.

The model scored 77.1% on ARC-AGI-2, a 2.5x improvement over Gemini 3 Pro's 31.1%, demonstrating a massive leap in genuine novel reasoning capability. On GPQA Diamond it achieves 94.3% (vs Claude Opus 4.6 at 91.3% and GPT-5.2 at 93.2%), and it launched at first place on both SWE-Bench Verified (80.6%) and Terminal-Bench 2.0 (68.5%). GDPval-AA sits at 1317 Elo, lower than the Claude models but still competitive.

Gemini 3.1 Pro offers three thinking levels — Low, Medium, and High — for controlling reasoning depth and compute cost. Pricing is tiered: $2/M input and $12/M output for requests under 200K tokens, scaling to $4/$18 for 200K-1M token requests. Context caching can deliver up to 75% cost savings on repeated context. A specialized gemini-3.1-pro-preview-customtools variant is available for agentic workflows.

The multimodal capabilities enable workflows that combine code review with screenshot analysis, document processing with image understanding, and video analysis with text generation — all in a single model. Its combination of top-tier reasoning benchmarks, native multimodality, competitive pricing with context caching, and the broadest platform availability makes it a strong contender for research analysis, long document processing, and agentic workflows.