Working on something…

GPT-6.1 Sol unusually often does commands or scripts that don't work. What is up with that? I don't remember 5.6 Sol doing that!

GPT-6.1 Sol is so stupid. The only reason to give money to @OpenAI now is to have unlimited access to AI data processing via GPT-6 Luna. And for that Plus with 20$/month is plenty.

GPT-6.1 Sol is still not available for me in Germany in the Codex desktop app on windows!!! @OpenAI @OpenAIDevs @sama @thsottiaux

500$ ChatGPT Sub is now out! With only 25% more usage than the original 200$ Plan :(

Perplexity can now run its Computer agent locally on Ryzen AI Max PCs PPLX 27B or Qwen 27B, local files stay on-device, and it only goes to the cloud with permission this hybrid local/cloud setup feels like a much better default for desktop agents (also in regards to limits) :)

Sonnet 5.5 is out!

GPT-6 Sol is the first model I’ve seen do this: It realized it got stopped by the 5 hour Limit and set up a loop to every hour, send it a continue message so that I don’t have to do that. Tecenologia

Codex really needs an Auto resume once limit resets function! @thsottiaux Especially for people with the 5h Limit!

Now would be a great time to remove the 5h Limit from Plus users! @thsottiaux

Opus 5.5 feels like a generational leap

TensorFold, a new engine that is used to serve a local LLM on Apple Silicon or NVIDIA GPU, is now being analyzed by me, Ash, and others, to compare performance vs. vLLM and SGLang. Current numbers are promising. If successful, there will be a new dawn of recipes for DGX Sparks and Apple Silicon. I would like to thank and congratulate @ashxhart on his work. Truly talented individual. Please follow him if you aren't already. github.com/ashhart/TensorFol…

TensorFold Inference Engine is here 🚀 I spent six months making one weight read count for more than one token on Apple Silicon. Draft tokens run through parallel lanes; the model verifies them together and keeps only what passes. Qwen 3.8 27B MLX 4Bit - 120-124tks Nemotron Lightning MLX 4Bit - 188-206tks Qwen3.8 Flash Next MLX 4Bit - 88-92tks CUDA Implementation is in Alpha showing strong gains. The Repo is in the comments 👇🏼

After Opus 5.5, I think it's time to retire "vibe coding." This is just how programming works now.

Any way that we could get some sort of an (manual) compact feature in @ChatGPT web (Chats)? @thsottiaux

99% of people are completely unaware of what AI can do RIGHT NOW they probably can't even imagine what AI is going to be able to do in say 1, 2 or 5 years from now

30min of GPT-6 Sol usage is about 50% of the 5h Limit and 8% of the weekly Limit on Plus.

Seeing only 3% 5h Usage gone after 30min of GPT-6 Luna in a /goal. That would mean that you could run 3 Luna Agents simultaneously 24/7 on PLUS!!!

Space Bunny might be the first actually really good and USABLE stealth model.

God Damn, even the non Fast mode of Opus 5.5 is really fast at 73 Tok/s. Fast mode ups this to almost 160 Tok/s!

Tether just released a 191B-token synthetic STEM dataset aimed at making smaller models reason better I like this direction a lot more than just throwing more parameters at everything better data is one of the biggest remaining levers for making actually useful local models