o3 and o4 Mini

Author

Aaron Samuel
April 21, 2025

Hey clever people,

This is the most important OpenAI update yet, 

Read this no-fluff summary

o3 and o4 Mini: OpenAI’s Next-Gen AI Models Are Here

Imagine an AI with a human-like brain. Well, OpenAI just launched o3 and o4 Mini, the newest models in its “O-series” lineup. They're built to think longer before responding.

Basically, these models are smarter, more agentic, and better at reasoning than anything OpenAI has released before. (Basically, they're like human brains without emotions. Imagine the power)

1. What’s the Hype? Key Features of the New Update

Advanced Reasoning: Both o3 and o4 Mini are optimized for complex, multi-step thinking, especially in:

  • Coding: Top scores on coding benchmarks

  • Math: Step-by-step explanations for tough problems 

  • Science & Logic: Strong at synthesis and analysis

  • Visual Reasoning: Can think with images, not just describe them 

2. All Tools, One Model

These models can use every ChatGPT tool, & know when and how to use them: (do you feel this update is smarter than your highly caffeinated brain?) 

  • Web browsing

  • Python for file analysis

  • Image generation

  • Code execution

  • Data visualization

  • And more

Trained with reinforcement learning to decide which tool to use for each task. This means more autonomy, less babysitting. 

3. Multimodal Thinking

They don’t just analyze images, they reason with them: (debating in AI) 

  • Understand diagrams, whiteboards, and sketches

  • Extract + integrate visual + text info

  • Zoom, rotate, and process blurry images

  • Blend visual and textual inputs into decision-making

O3 vs. O4 Mini: What’s the Difference?

Feature

O3

O4 Mini

Power

Most powerful OpenAI model to date

Lightweight, efficient, and fast

Strengths

Best at reasoning, coding, consulting, and creative thinking

Great for math, coding, and image tasks at scale

Accuracy

20% fewer major errors vs O1

Occasionally outperforms O3 in specific benchmarks

Use Case

Deep problem-solving

High-volume, cost-effective reasoning

Cost (API)

$10 input / $40 output per million tokens

$1.10 input / $4.40 output per million tokens

Bonus: There’s an O4 Mini High version for heavier reasoning tasks.(There is always a bonus with AI updates) 

IRL Examples

  • Solve complex, multi-step business or technical problems

  • Read and reason with visuals + code + text in one prompt

  • Run end-to-end logic with tool orchestration

  • Debug and write cleaner and faster code

  • Extract info from blurry charts, annotated whiteboards, or old PDFs

Codex CLI – For Devs Who Live in the Terminal

OpenAI also dropped Codex CLI, an open-source agent you can run straight from your terminal. 

Built for O3 and O4 Mini

  • Supports multimodal reasoning from the command line

  • OpenAI is offering $1M in grants to devs using it ($25k each)

Access + Availability

  • Plus, Pro, and Team users: Full access to O3, O4 Mini, and O4 Mini High

  • Enterprise + EDU: Rolling out within a week

  • Free users: Can test O4 Mini by selecting “Think” in ChatGPT

  • Developers: Available via Chat Completions + Responses API

More AI updates are landing in your inbox soon.

— Written by Aaron & The Clever Nest Team

If you found this helpful, forward it to your co-founder who still thinks GPT-3.5 is “good enough.”