Hey friend,

We have a lot of new AI model releases to cover this week. Today, we’ll explore

• Google's Gemini 2.5 Flash model

• OpenAI’s latest releases (o3, o4-Mini, GPT-4.1)

• How to transcribe YouTube videos with AI

• A killer new AI feature in Google Sheets

Let's dive in.

Tutorial: How To Transcribe YouTube Videos With AI

This week I wanted to share a trick I’ve been using to transcribe YouTube content using Google Gemini which is a powerful multi-model model. Historically most people have relied on paid third party tools and services to do this. I’ll walk you through step-by-step how to automate YouTube transcript generation with Gemini and N8N. Reply to this email if you have any requests for tutorials.

AI News

Here’s my pick of AI news this week:

1. Google's Gemini 2.5 Flash With 'Thinking Budget'

Google has launched Gemini 2.5 Flash, a multimodal AI model featuring innovative hybrid reasoning capabilities. It allows developers to toggle between "thinking" modes and set token-based budgets, effectively balancing performance, latency and cost for high-volume enterprise applications.

• The model's hybrid reasoning approach enables dynamic adjustment between deep processing ($3.50/1M output tokens) and simple query responses ($0.60/1M output tokens), with customisable "thinking budgets" to optimise costs based on task complexity.

• Gemini 2.5 Flash can process up to 1 million tokens of text, audio, images, or video (equivalent to 6 Harry Potter books), whilst maintaining speeds twice as fast as Gemini 2.5 Pro and improved coding and factual accuracy over 2.0 Flash.

2. OpenAI release o3 And o4-Mini

OpenAI has launched o3 and o4-mini, advanced reasoning-optimised AI models designed for complex multimodal problem solving. The models integrate web browsing, code execution, and image analysis tools, establishing new benchmarks in AI agents' ability to solve multi-step problems independently across coding and visual reasoning tasks.

• o3 reduces major errors by 20% compared to o1 on difficult real-world tasks, achieving 100% accuracy on AIME 2025 with Python tools, while o4-mini delivers 99.5% accuracy at 80% lower computational cost than o3.

• The models are immediately available to ChatGPT Plus/Pro/Team users.

3. OpenAI's Codex CLI Brings Secure AI Coding To Your Terminal

OpenAI has launched Codex CLI, an open-source terminal tool that uses its newest AI models (o3 and o4-mini). The tool enables natural language coding, debugging and file operations directly in the command line whilst keeping data on-device for enhanced security. You can find it on Github.

This is similar to Claude code a similar terminal based AI coding assistant.

• Codex CLI runs natively on macOS/Linux with experimental Windows support through WSL, and accepts multimodal inputs including text prompts, screenshots and hand-drawn sketches combined with local code context for enhanced reasoning.

• The tool processes code locally without cloud dependencies, addressing developer concerns about sensitive data exposure common in web-based AI coding assistants, with future plans for CI/CD pipeline integration and expanded team workflow features.

4. OpenAI's GPT-4.1 Boosts Coding Accuracy By 21% With Million-Token Context

OpenAI has launched three new API-exclusive models—GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano—featuring 21-27% coding improvements over predecessors, million-token context windows, and enhanced instruction following. These models replace GPT-4.5 Preview and target developers building AI agents for software engineering and enterprise applications.

• The models achieved 54.6% accuracy on SWE-bench coding tasks (21.4% better than GPT-4o) and 72% on Video-MME long-context understanding, demonstrating significant performance improvements in technical capabilities.

• With 1-million-token context windows, the models can process documents containing over 750,000 words while maintaining improved positional awareness throughout the entire context length.

• OpenAI's API-only approach ensures enterprise-grade reliability, with GPT-4.1 offering 30% cost reduction per API call compared to GPT-4.5 Preview, which will be deprecated on 14 July 2025.

Tutorial: Google Sheets “=AI” Formula

Google has a killer spreadsheet feature that allows you to use AI as a formula. Most people don't know how powerful it is.

To use it:

1. Go to Gemini for Google Workspace settings

2. Enable "Alpha Features"

3. Go to Google Sheets

4. Use the formula '=AI'

You can enter a specific prompt to describe your desired action and add an optional range: =AI("prompt", [optional range])

Your Opinion Matters

What did you think of today’s email? Your feedback helps me create better emails for you!

Got more feedback or want me to cover a specific topic? Reply to this email and let me know.

Owain