NEW DeepSeek V3.2 Thinking: Equal to Gemini 3 PRO?

Finally. The improved version of DeepSeek 3.2 was released today … With maximum Test Time Scaling. I am testing the thinking and non-thinking version live. The experimental version (now outdated) of DeepSeek 3.2 EXP was released 2 months ago.

Will this new Chinese AI perform at the same level like GPT 5 (high) or even Gemini 3 PRO? We look at the official marketing benchmark data and perform a real world live test with both models, on the Chinese and the US platform. The results are different than the benchmarks.

https://www.deepseek.com/en

@deepseek-api @googledeepmind

00:00 DeepSeek V3.2 Thinking benchmarks
03:18 Live Test DeepSeek V3.2 Thinking
06:08 Test DeepSeek 3.2 non-thinking
06:40 Test on LMarena platform
09:39 Final thoughts on DeepSeek 3.2

#aitesting
#aitest
#airesearch
#aireasoning
#deepseek

source