Meet o3-mini, OpenAI’s Efficient Thinker

OpenAI delivers on its promise and drops o3-mini in January…

In this edition we’ll be covering…

• OpenAI’s o3-mini model and how to access it (for free)

• Deep Research from OpenAI

• Our guide on how to use Phi-4 locally

• 5 hot AI topics

• 3 tools to keep you more productive

• And much more…

The Latest in AI

Meet o3-mini, OpenAI’s Efficient Thinker

Smarter AI just got even faster.

On Friday, OpenAI dropped o3-mini, the latest addition to its reasoning model series.

Think of it as o1's younger sibling who went to business school – same smarts, better with money…

The breakdown:

• o3-mini delivers comparable results to its predecessor o1 in mathematics, coding, and science while optimizing for efficiency.

• Users can toggle between low, medium, and high reasoning effort.

• It supports function calling, structured outputs, developer messages, and streaming, plus beta search capabilities for real-time information retrieval.

• The model is available to ChatGPT Plus, Team, and Pro users, with free-tier access via the Reason option. API pricing is significantly lower than o1.

So What?

With DeepSeek making waves, OpenAI is pushing back with a more accessible, cost-effective model. Plus, the addition of reasoning-backed search in ChatGPT? That’s a game-changer.

To start using o3-mini:

1. Log into ChatGPT. In Pro and Plus plans, you can select it directly in the model dropdown.

2. To access it in the free tier, select the Reason button under the chat window.

3. Enter your question to get reason-driven results!

Together with Fellow

The #1 AI Meeting Assistant

Typing manual meeting notes drains your energy. Let AI handle the tedious work so you can focus on the important stuff.

Fellow is the AI meeting assistant that:

✔️ Auto-joins your Zoom, Google Meet, and Teams calls to take notes for you.
✔️ Tracks action items and decisions so nothing falls through the cracks.
✔️ Answers questions about meetings and searches through your transcripts, like ChatGPT

Try Fellow today and get unlimited AI meeting notes for 30 days.

Get Started Free

Get Your Hands Dirty

Run AI Locally with Microsoft’s Phi-4

Speaking of small models, Microsoft officially released Phi-4 last month—a 14-billion parameter, state-of-the-art language model optimized for a variety of tasks.

The best part? It’s open-source, meaning you can run it locally for free without worrying about your data being sent to external servers.

Here’s how you can use it with Ollama:

1. Download and install Ollama

2. Pull the model:

ollama pull phi4

3. Run the model locally and start asking questions!

ollama run phi4

Now that you have this model installed on your computer, you can build cool AI apps locally.

Note: Before you get too excited, check your hardware. If your computer struggles with Chrome tabs, it might not be ready for Phi-4. For reference, I run 8B parameter models on an M2 MacBook Pro, and Phi-4 is a bit beefier at 14B parameters.

Industry Intel

ChatGPT Gets a Significant Research Upgrade

OpenAI just launched Deep Research in ChatGPT—a new agentic capability designed to conduct multi-step research across the web, condensing hours of human work into minutes.

What’s cooking…

• Deep Research searches, analyzes, and synthesizes information from hundreds of sources to generate comprehensive reports, going beyond just traditional search.

• Unlike real-time chat with plain GPT-4o, Deep Research excels in domain-specific, in-depth inquiries, making it ideal for finance, science, policy, and shopping research.

• The capability has been rolled out to ChatGPT Pro users at 100 queries per month so far, and Plus users will gain access next.

• To use it, select Deep Research under the message box in ChatGPT, then ask away!

So What?

OpenAI just turned ChatGPT into a 24/7 research department. Whether you need competitive analysis, scientific deep dives, or policy insights, Deep Research acts like a team of PhD students at your fingertips—without the coffee budget.

Even Sam Altman claims Deep Research can handle a single-digit percentage of all economically valuable tasks worldwide.

We recently covered Google Gemini’s Deep Research, and you better believe we’ll be doing a full comparison soon 👀.

Quick Bites

Stay updated with our favorite highlights, dive in for a full flavor of the coverage!

A new paper from the Anthropic Safeguards Research Team describes a method that defends AI models against universal jailbreaks.

AI-generated content has been suspected of overwhelming many online and social media platforms such as LinkedIn, Reddit, and even Google. Check out this study done by Originality.ai.

DeepSeek-R1 seems to be failing every safety test thrown at it when tested with malicious attacks.

LangChain breaks down o3-mini’s capabilities, use cases, and how it compares to other models in the O-series family.

Google announced "Ask for me," an experiment in its Search Labs testing ground for Google Search. The feature uses AI to call local businesses on your behalf and ask about pricing and availability.

NVIDIA CEO Jensen Huang says everyone should get an AI tutor right away.

Trending Tools

🧠 Mindsmith - An authoring tool that applies AI to every step of the eLearning creation process.

🫂 Groops - A lead-generation system that uses AI to create landing pages.

📖 SoBrief - Read any book in 10 minutes with AI.

The Neural Network

Just replace this with o3 and this becomes extremely accurate…

Until we Type Again…

Thank you for reading yet another edition of Digestible AI!

If you have any suggestions or specific feedback, simply reply to this email. Additionally, if you found this insightful, don't hesitate to engage with us on our socials and forward this over to your friends!