Is OpenAI really in trouble?
Don't listen to DeepSeek hypebros
The newsletter for the technically curious. Updates, tool reviews, and lay of the land from an exited founder turned investor and forever tinkerer.
Hey folks,
The Information reports Sam Altman declared “Code Red” inside OpenAI after Gemini 3’s success. Ads in ChatGPT will be delayed, with the same fate for products like Agents and Pulse. Personalisation, image generation and model behaviour will be the main focus. The report also claims that OpenAI will release a new reasoning model better than Gemini 3 next week.
DeepSeek’s new models, DeepSeek V3.2 base and Speciale, are impressive (and open) BUT not better than any of the other top models (GPT-5.1/Codex, Opus/Sonnet 4.5, Gemini 3). The Speciale version can get a gold medal in math olympiads, but it’s only good for proving theorems, it seems, and you can think of the base version as almost the same level as GPT-5 though you’ll see subpar results in both use cases: chat and coding.
FLUX models by Black Forest Labs were the original stable diffusion killer. With FLUX.2 family of image generation models, their next goal is Nano Banana and dozens of open models from China. FLUX.2 [dev] is the 32B open weights variant, better and 3x cheaper than the original Nano Banana (10x cheaper than Nano Banana Pro). BFL also raised a $300M Series B.
The leading video generation model swapped hands by a significant margin. Runway is the new leader with its latest Runway Gen 4.5 model, beating Google’s Veo 3. Kling, a close competitor, also released a new model, Kling O1, exclusively on Fal, so it’s not on the leaderboards yet.
Attio is the AI-native CRM for the next generation of teams. Sync your email and calendar, and Attio instantly builds your CRM—enriching every company, contact, and interaction with actionable insights in seconds. Join fast growing teams like Granola, Flatfile, Modal, and more. Start for free today.*
🌐 What I’m consuming
The new alignment research blog from OpenAI, live with its first two posts.
On the consumption of AI-generated content at scale.
Thoughts from building and shutting down a portable context layer company.
OpenAI’s lead under pressure as rivals start to close the gap - FT
Supermemory is state-of-the-art on LongMemEval - a measure for reasoning, recollection and more across 100K+ tokens. (I’m an investor)
⚙️ Tools and demos
Concierge.ai - Engage & convert inbound website visitors into qualified leads with a custom AI answer engine trained on your brand & content.*
Gemini Dynamic View - Paid users on Gemini can select this tool to get a visual one-pager website-like output on a certain topic. Here’s a sample of using it to catch up for Stranger Things S5.
Duet - Group ChatGPT for work (with MCP and a shared knowledge base).
Rephrase - Fix grammar and format text instantly on Mac.
Komposo - Generate and edit designs, and export directly to code.
Plok.sh - Turn a GitHub repo into a blog. Not an AI tool, but this looks useful.
TLDW - Paste the link of an hour-long video and get the gist in 5-minute highlight reels.
🥣 Dev dish
osgrep - open source, local semantic code search for Claude Code that works.
WarpGrep by Morph runs parallel tool calls as a context subagent to improve coding agent performance.
Lux by OpenAGI - Fast, cheap and better computer-use model in an SDK.
vibeproxy - Native macOS menu bar app to use your Claude/ChatGPT/Gemini subscriptions with AI coding tools.
🍦 Afters
Valon is hiring Forward Deployed Engineers. $130K–$230K & equity, turning enterprise clients needs into code onsite. NYC/SF/Seattle + travel*
Raindrop AI (monitoring and A/B testing for AI apps) raised $15M seed.
Telegram is operating a decentralised compute network for its AI features and making it open for others, too.
OpenAI takes an ownership stake in Thrive Holdings.
Thrive Holdings (initial round of $1B) acquires and holds legacy businesses like accounting & IT services and transforms them with AI/tech. OpenAI now has equity and will train models for tasks with company-specific data.
- from Sheel Mohnot’s tweet (@pitdesi)
That’s it for today. Feel free to comment and share your thoughts. 👋
Read about me and Ben’s Bites
📷 thumbnail creds: @keshavatearth,
Thanks to today’s sponsors who made this newsletter possible :)
Wanna partner with us? Last few slots left for the rest of the year.
Purely N of 1 data, but I accidentally started using gemini 3 while GPT was grinding out something. I haven't used GPT since. I find that for a researcher the quality was much greater. I dropped my $200 a month sub the next week. (Im a 10 hour a day user.) N of 1. Still.
Love this!