分页： 1 / 1
#1 Deepseek r1 vs open ai o3 on arc-agi tests
发表于 ： 2025年 2月 5日 19:06
由 itspid
According to recent analysis, DeepSeek's "R1" model achieved a score of around 15% on the ARC-AGI benchmark, with its "R1-Zero" variant scoring slightly lower at approximately 14%. This means that while DeepSeek models show promising reasoning abilities, they still fall significantly below the scores achieved by top-performing AI systems like OpenAI's "o3" which scored around 87.5% on the same test.
Deepseek r1 比新出的o3差得远去了，是不是要想办法再蒸一遍模型。
#2 Re: Deepseek r1 vs open ai o3 on arc-agi tests
发表于 ： 2025年 2月 5日 19:42
由 TheMatrix
itspid 写了： 2025年 2月 5日 19:06
According to recent analysis, DeepSeek's "R1" model achieved a score of around 15% on the ARC-AGI benchmark, with its "R1-Zero" variant scoring slightly lower at approximately 14%. This means that while DeepSeek models show promising reasoning abilities, they still fall significantly below the scores achieved by top-performing AI systems like OpenAI's "o3" which scored around 87.5% on the same test.
Deepseek r1 比新出的o3差得远去了，是不是要想办法再蒸一遍模型。
o3到目前为止谁也没见过，都是自己说。
#3 Re: Deepseek r1 vs open ai o3 on arc-agi tests
发表于 ： 2025年 2月 5日 23:19
由 Caravel
TheMatrix 写了： 2025年 2月 5日 19:42
o3到目前为止谁也没见过，都是自己说。
这些新的benchmark先澄清一下有没有oai的投资
#4 Re: Deepseek r1 vs open ai o3 on arc-agi tests
发表于 ： 2025年 2月 6日 11:35
由 赖美豪中
不用担心，只要发布了2小时就能对标。我实测过ds错了gpt对的，基本2小时到一天内都会和gpt的答案一致，我估计后来有人工实测校正
itspid 写了： 2025年 2月 5日 19:06
According to recent analysis, DeepSeek's "R1" model achieved a score of around 15% on the ARC-AGI benchmark, with its "R1-Zero" variant scoring slightly lower at approximately 14%. This means that while DeepSeek models show promising reasoning abilities, they still fall significantly below the scores achieved by top-performing AI systems like OpenAI's "o3" which scored around 87.5% on the same test.
Deepseek r1 比新出的o3差得远去了，是不是要想办法再蒸一遍模型。