12 hours ago Deep Research was unveiled, and I’ve tested it thoroughly, including vs Deepseek R1 with search, Gemini Deep Research and even R1 in Perplexity. It’s a notable step forward, with one big caveat. I’ll go through all the benchmark figures, my initial impression of the o3 model within, and much more.

Deep Research: https://openai.com/index/introducing-deep-research/

https://www.youtube.com/watch?v=YkCDVn3_wiw

GAIA Bench: https://openreview.net/forum?id=fibxvahvs3

https://openreview.net/pdf?id=fibxvahvs3

CodeELO:https://arxiv.org/pdf/2501.01257

CamelCamel:https://uk.camelcamelcamel.com/

Deepseek R1 with search: https://chat.deepseek.com/

https://arxiv.org/pdf/2501.12948

HaluBench: https://arxiv.org/pdf/2407.08488

Chapters:

00:00 - Introduction

01:06 - Powered by o3, Humanity’s Last Exam, GAIA

03:55 - Simple Tests 

06:00 - Good News vs Deepseek R1 and Gemini Deep Research

09:32 - Bad News on Hallucinations 

14:14 - What Can’t it Browse?

14:42 - For Shopping?

16:40 - Final thoughts