OpenAI's roadmap update - GPT 4.5 & GPT 5

Hey there!

Welcome back to The Pulse, where we dive into interesting AI stories and trends backed by data, all presented through simple visuals.

Top news pick: Everyone’s heading towards deep research

Over the last three months, three big AI companies - OpenAI, Perplexity AI, & Google - have announced their own versions of deep research. It is an AI research technology that can conduct multiple web search to produce comprehensive reports on any topic. They’re all powered by the respective state-of-the-art AI models. Talk about having a burden of choice. 😅 ‍

For more news like this, check out 42neurons!

> GPT-4.5 to be last non-COT model; GPT-5 to integrate o3 + other OpenAI tech

> no more model picker, ChatGPT to decide how long to think based on tasks

> ETA for models according to Sam Altman: weeks/months from now

> responds to queries at speed of 1,100 tokens/s through Flash answers feature

> powered by Cerebras Inference and 123B-parameter Mistral Large 2 model

> ~10x faster than other models, eg. GPT 4o (115 t/s), Claude 3.5 (71 t/s), o1-mini (177 t/s)

> not as accurate as other slower models like DeepSeek R1

> most engagement in GenAI mobile apps go to AI companionship apps

> major companionship apps: Character.ai, Replika, etc.

> avg ~2 hours a day spent on Character.ai + over 200M visits every month

.