📄 中文摘要
Google 最近发布了 Gemini 3.1 Pro，尽管科技界对其令人印象深刻的基准分数热议不已，但最引人注目的细节并不在市场宣传中，而是在模型卡的第八页。Gemini 3.1 Pro 在纸面上表现出色，ARC-AGI-2 的得分高达 77.1%，在复杂推理任务如 GPQA Diamond 和 LiveCodeBench 中表现优异。这次更新解决了之前一个异常情况，即 'Flash' 版本的模型实际上超越了其他版本，标志着编码能力和逻辑推理的显著提升。
Gemini 3.1 Pro：超越基准测试与人工智能情境意识的崛起
📄 English Summary
Gemini 3.1 Pro: Beyond Benchmarks and the Rise of AI Situational Awareness
Google has recently released Gemini 3.1 Pro, and while the tech community is abuzz about its impressive benchmark scores, the most intriguing details lie not in the marketing materials but on page 8 of the model card. On paper, Gemini 3.1 Pro is a powerhouse, achieving a remarkable 77.1% on ARC-AGI-2 and excelling in complex reasoning tasks such as GPQA Diamond and LiveCodeBench. This update addresses a previous anomaly where the 'Flash' version of the model was actually outperforming others, indicating a significant leap in coding proficiency and logical deduction.