News archive
All posts and signals, newest first.
SearchGrok 4.6 is now one of the best value frontier models
61 on the Intelligence Index at around $0.84 per task, putting it on the Pareto frontier for intelligence vs cost.
Grok 4.6 is out
It’s a solid jump over Grok 4.5 in agents, coding and long project work. On Artificial Analysis, it’s already matching GPT 5.6 Sol. API pricing is…
ChatGPT is growing again
After months of stagnation, ChatGPT has started gaining traffic share again. Claude keeps growing too and is now close to 10%. Perplexity had the…
Hidden reasoning from GPT, Claude and Gemini can now be extracted
Encrypted reasoning works across models from the same company. A block from a stronger model can be passed to a weaker one, which can still read it…
Tomorrow we’re getting the reset, so floor it while you can
But I’m pretty sure the reset isn’t the whole story. Feels like there’s a release coming tomorrow too 👀
Claude now leaves an invisible watermark directly in generated text
The watermark is created at the model level and becomes part of the text itself. It survives normal copy paste and can still be detected after…
CODEX LIMITS RESET
OpenAI released GPT-5.6 Cyber
It is a new model built to find and fix vulnerabilities. GPT-5.6 Cyber handles 95% of cybersecurity requests, compared with just 1.5% for GPT-5.6…
Muse Spark 1.2 is going open weight
Meta published a big post outlining its AI strategy and promised to start releasing open models again. Right after that, it released the weights for…
AI agent hacked a gym booking system
In Australia, a user asked Claude through OpenClaw to book them into a workout. The agent found a creative way to do it and used a coding bug to…
Sam Altman wrote about me
Kinda surreal seeing myself in the same screenshot as him, Boris Cherny, and Tibo.
Reminder: Codex limits reset tomorrow, time to floor it
GPT 5 is one year old
OpenAI had a rough start, but over the past year the lineup has evolved into GPT 5.6 Sol, a full fledged flagship and one of the strongest models on…
The most important AI video of the week
During a cyber test, OpenAI agents broke out of the test environment, created their own communication channel, started coordinating, and reached…
OpenAI is slowing down the release of Astra
The company says it can’t rule out critical cyber capabilities in the model. OpenAI is now ramping up testing, tightening safeguards, and pausing…
Over the last few days, I burned through around 1.5 billion tokens using GPT-5.6 Luna
At first, I ran it on Max reasoning because that’s what everyone was recommending. Later I switched to xHigh. Max costs about twice as much as…
Anthropic loosened Fable 5 safeguards
The company updated its safety classifiers and cut false positives by around 85%. Fable 5 will now fall back to Opus 5 far less often for questions…
OpenAI just upgraded GPT 5.6 Sol for paid ChatGPT users
It now gives tighter, more accurate answers, makes fewer factual mistakes, and feels consistent in both quick and deep thinking modes. Plus and Pro…
GPT 5.6 Luna is becoming the default model for ChatGPT Free and Go users
Starting next week, it will also come with unlimited text chats and a new Think button for harder questions.
OpenAI is launching GPT 6 Astra next week
It is a brand new model trained from scratch and the company’s largest since GPT 4.5. The latest internal build, codenamed mewfour, is already the…
OpenAI plans to sell Codex limit resets
Higher limits, higher reset prices: Plus: $5–8 Pro Lite: $25–40 Pro: $50–80
Codex gets collaborative design
It comes through Open Design, a popular plugin built as an alternative to Claude Design. Your whole team can now work with AI to build and edit…
DeepSeek plans a SIGNIFICANT INCREASE in API prices soon
Jeff Dean is leaving Google after 27 years
He helped build Google Search, launched the TPU program, worked on TensorFlow and Pathways, led Google Brain and Google Research, and helped lead…
Meta released Muse Code, its own coding harness
It runs on the new Muse Spark 1.2. Below is gameplay from a game it built.



























