Benchmarking Gemini 3.1 Pro: Latency, cost, and reasoning trade-offs

Google's Gemini 3.1 Pro, announced in February 2026, represents a significant advancement in AI models, particularly for developers who require superior reasoning capabilities without increased costs. This model outperforms its predecessor and rivals, achieving noteworthy scores on benchmarks like ARC-AGI-2 and surpassing models from Anthropic and OpenAI. It offers adjustable thinking levels—low for speed, medium for balance, and high for full reasoning—allowing developers to control response times and accuracy based on task complexity. Despite its enhanced performance, Gemini 3.1 Pro maintains competitive pricing with unchanged rates from previous versions, and its tiered token billing positions it as a cost-effective alternative to competitors like OpenAI. The model's strengths lie in complex analytical work, though it doesn't claim universal dominance, and its flexibility is highlighted by multimodal support and a large context window. By providing tunable reasoning depths and maintaining stable pricing, Google empowers developers to optimize their applications based on specific needs, making Gemini 3.1 Pro a versatile and valuable tool in the AI landscape.