o3 is now 5x cheaper
Veo 3 killer, Zuck's new hire and Mistral reasoning
I write a newsletter about startups and investing—for ai builders of all levels.
I record mini-tutorials, review tools I’m testing, share my insights from an exited founder turned investor.
Hey folks,
Twins bday today so keeping the intro light!
Told you it was light.
🔎 News worth knowing
OpenAI launched o3 Pro for Pro and (with limits) Teams plans. o3 pro feels much better than o3, but the thinking time for each query makes it an annoying model to use. I’m getting 10-15 mins of thinking time per query on average, and when the code doesn’t work after 20 mins of scrolling twitter, it’s frustrating. Sam Altman also confirmed the 80% price cut for o3, making it now cheaper than GPT-4o in API (great time to be a builder).
ps: Sam has a new essay with some stats on how much energy a single ChatGPT query consumes.
Mistral has released its first batch of reasoning models, Magistral Small (open source) and Magistral Medium (for enterprises). They aren't the best in their weight class, but Mistral's selling point is that they are the only usable models from a European lab. They are also selling a new AI infrastructure offering to train and serve AI models for enterprises, called Mistral Compute.
ByteDance (TikTok’s parent co) has released a new video generation model called SeedDance that they claim is better than Veo-3. They released a new video generation benchmark and an arena as well.
Build privacy-first Voice AI Agents. No need to compromise accuracy for security. Speechmatics lets you deploy enterprise-grade AI Voice Agents on-prem or on-device, keeping customer data private while handling noisy, real-world speech across 55+ languages. Try Speechmatics for free.*
Meta is reportedly building a new superintelligence team, for which it's paying roughly $15 billion to acquire a 49% stake in Scale AI and bring on Scale's CEO, Alexandr Wang, as a top Meta AI executive. What will Meta gain from this?
*sponsored
want to partner with us? Click here
🌐 What I’m consuming
16 changes for AI in the enterprise—spending is growing and becoming permanent, and enterprises are testing and using multiple models.
How to prompt Veo-3 for the best results.
How Vercel is adapting SEO for LLMs and AI search.
How Intercom is building back using ai
How a founder used AI to save his company from a two-year litigation nightmare.
In consumer AI, momentum is the moat.
Using claude code to ship like a team of 15—when you’re only a duo.
Cursor’s team is on a podcast run. I see them everywhere, but this video with Anthropic is a good watch. I also highly recommend Ben Thompson’s interview with one of the co-founders, Michael.
This professor is teaching national security and letting his students use AI. The blog has examples of student groups using AI and the professor's comments.
The team at Every made LLMs compete in a game of Diplomacy. o3 and Gemini 2.5 Pro are the big dogs.
⚙️ Tools I’m tinkering with
Amplitude AI agents for product managers to build self-improving products.
Scouts are always-on AI agents that monitor the web for news, price drops, or flights.
Replit now allows users to import a) their Figma designs and b) their apps from Bolt or Lovable.
Windsurf is introducing a planning mode and a browser for easy debugging.
Claude Code is also getting a plan mode that allows you to review Claude’s plan before it starts making changes to your files.
Dia, the new AI browser from the makers of Arc, is launching in beta. (article in the verge) - I loved Arc until they gave up on it :(
🥣 dev dish
LightRT-LM, a C++ library from Google to run open-source models on-device.
FireEnrich, an open-source Clay alternative to enrich people and company data.
Someone built an OpenAI Codex clone using VibeKit.
Fine-tuning for GPT 4.1 with DPO is now available in OpenAI playground.
🍦 Afters
Supabase is hiring for their new product to scale Postgres.
Salesforce is blocking AI rivals from using Slack data, like the interactions users are having with Glean's Slack chatbot.
OpenAI is now buying cloud compute from Google Cloud, some of which is coming via CoreWeave.
That’s it for today. Feel free to hit reply and share your thoughts. 👋
Enjoy this newsletter? Please forward to a friend.