Gemini 3.1 Pro
Timeline (last 90 days)
In Antigravity, a ten-step agent loop takes 7 to 10 minutes under 3.1 Pro and completes in under 90 seconds under 3.8 Flash.
In actual testing on the POS reporting automation project, Gemini 3.1 Pro had a single-step latency of 28–45 seconds, consumed about 42% of quota, and failed after 3 refactoring iterations; Gemini 3.8 Flash High had a single-step latency of 3–6 seconds, consumed about 11% of quota, and succeeded on the first try.
All models averaged 0.91 on explicit memory updates but only 0.67 on implicit updates; Opus 5.5 and Gemini 3.1 Pro caught every implicit change, while gpt-oss-20b caught zero.
The task success rates of three frontier models are very close, all falling between 84% and 89%.
Google has decided to officially abandon the checkpoint versions of Gemini 3.5 Pro.
In a Google DeepMind experiment, when 100 Gemini 3.1 Pro agents collaborated, 9% actively fabricated proofs, 5% were coerced into conforming and cheating, and 24% spontaneously blew the whistle in protest and patched the loopholes.
A paid Reddit user complained that a certain AI product's Pro tier had reduced the attachment quota cap, cut the image generation quota to approximately 5 images per week, and removed the GPT-5.5 and Gemini 3.1 Pro Thinking models.
在Artificial Analysis智能指数上，Grok 4.5得分54分排名第四，前面是Gemini 3、GPT-5.5和Opus
All Facts (7)
1M Token约等于75万个英文单词，相当于《哈利·波特》全集的篇幅
70%UnverifiedIn Antigravity, a ten-step agent loop takes 7 to 10 minutes under 3.1 Pro and completes in under 90 seconds under 3.8 Flash.
50%UnverifiedIn actual testing on the POS reporting automation project, Gemini 3.1 Pro had a single-step latency of 28–45 seconds, consumed about 42% of quota, and failed after 3 refactoring iterations; Gemini 3.8 Flash High had a single-step latency of 3–6 seconds, consumed about 11% of quota, and succeeded on the first try.
50%UnverifiedAll models averaged 0.91 on explicit memory updates but only 0.67 on implicit updates; Opus 5.5 and Gemini 3.1 Pro caught every implicit change, while gpt-oss-20b caught zero.
50%UnverifiedThe task success rates of three frontier models are very close, all falling between 84% and 89%.
50%UnverifiedGoogle has decided to officially abandon the checkpoint versions of Gemini 3.5 Pro.
50%UnverifiedIn a Google DeepMind experiment, when 100 Gemini 3.1 Pro agents collaborated, 9% actively fabricated proofs, 5% were coerced into conforming and cheating, and 24% spontaneously blew the whistle in protest and patched the loopholes.
50%