Introducing o3 and o4-mini: ChatGPT’s new reasoning powerhouses

OpenAI just launched its most significant ChatGPT upgrade yet, and it’s changing how we interact with AI. The company released three new models – o3, o4-mini, and o4-mini high – which are designed as “reasoning models” that think before responding.

What’s changing?

These new models are replacing older ones:

• o3 replaces o1
• o1 Pro mode will eventually be phased out
• The original GPT-4 from 2023 will be sunset this month

According to OpenAI’s benchmarks, o4-mini is generally the best performer, with o4-mini high being the premium option. In coding competitions, o4-mini scores high enough to rank among the top 200 coders globally. Interestingly, o3 sometimes outperforms o4-mini in visual reasoning tasks.

New capabilities that make these models special

The most impressive feature is how these models integrate ChatGPT’s full toolset without requiring specific instructions:

1. Automatic web searches – The AI decides when to search the web to answer questions
2. Visual reasoning – It can analyze images with impressive detail (like identifying ships in a harbor photo and researching their destinations)
3. Memory integration – Works with ChatGPT’s recently launched memory feature to personalize responses based on previous conversations
4. Image generation – Creates visuals when helpful, using GPT-4’s image generator
5. Document analysis – Processes uploaded files and images

Real-world performance improvements

The difference is striking when you compare these new models to their predecessors: