BREAKING
OpenAI Launches o1-preview & o1-mini
Models That Take Time to Think
AIME: o1-preview 83% vs GPT-4o 13%
o1-preview
83
GPT-4o
13
0
%ile
Codeforces
0
/100
Safety test
0
k
Context
Strengths and Trade-offs
Strengths
●
Subtle code-review bugs
●
PhD-student level
Concerns
●
Overwhelming internal CoT
●
GPT-4o still practical
Inference-Time Compute in Focus
AI NEWS BLITZ
OpenAI has unveiled its new o1 model series, built for complex reasoning.