By
–
We've compared 4 leading coding models (best from each lab) with some fun SVG generations, see which one you like best. I'm honestly quite impressed, this is leaps and bounds better than only a few months ago. https://t.co/70NGIg2TH7
— Peter Gostev (@petergostev) 1 janvier 2026
We've compared 4 leading coding models (best from each lab) with some fun SVG generations, see which one you like best. I'm honestly quite impressed, this is leaps and bounds better than only a few months ago.
By
–
Once again people don't understand the exponentials. In 10 years time there will be nothing but epoch plots
By
–
We need a word for the loss you experience when you hit context length limit and the state of the LLM that you were on the same wavelength for so long gets reset and you might never get it back again
By
–
Having used both Claude 4.5 Opus and GPT-5.2-xhigh side by side I can say that Opus has a beautiful mind, it's like an artisan who can come up with wonderful ideas & quickly. But if I don't want the model to suddenly flip the table and kick me in the nuts, I switch to GPT-5.2,
By
–
Frontier AI data center capacity, based on data collected by @EpochAIResearch. It doesn't include every single one, as they focus on the largest ones.
— Peter Gostev (@petergostev) 29 décembre 2025
Few things stand out:
– 2026 will have a huge amount of cpacity come online
– Anthropic will lead at some early points in… pic.twitter.com/lJfxjMc87J
Frontier AI data center capacity, based on data collected by @EpochAIResearch
. It doesn't include every single one, as they focus on the largest ones. Few things stand out: – 2026 will have a huge amount of cpacity come online – Anthropic will lead at some early points in
By
–
A thought, would it be worth having a 'side chat' to your codex agent while it is working away? I know you can have multiple terminals at the same time but it is a bit more complex to manage. If it is working for 20 mins, I wouldn't mind speaking to a sub agent to ask it stuff
By
–
The best early application of continuous learning is your personal LLM-as-a-judge. Imagine an LLM that knows exactly how you code, write emails, what you consider to be 'good' legal document or what is a good design – anything that you normally give feedback to LLMs on is
By
–
Do you remember how people losing their minds over Sonnet 3.5 for coding? What innocent times
By
–
My wish is that we get reasoning music models, but didn't put that into predictions because I doubt this will actually happen
By
–
Here are my 26 predictions for 2026. I tried hard to come up with more spicy predictions that are still plausible – so they sit somewhere in the 5-60% range for me. China 1. Chinese open model tops the Web Dev Arena for 1+ months 2. Chinese labs will open source less than 50%