Timothée Chauvin

(lmk if there's a better way than /usage, especially suitable for a background process, because that command is actually quite unreliable)

some tips if you're poor: you can (currently) get free Opus 4.5 credits in Antigravity, with higher limits if you have Google AI Pro (which students get free for 1yr). You can also run claude /usage at strategic times to start a new window, because they set their 5-hour windows adversarially

I'm interested in UIs + models that have minimal latency, please share suggestions!

YouTube video by Timothée Chauvin Latency of a few LLMs and google search on an easy question

...making me lose the cursor and forcing me to use the mouse, as you can see in the video.

Link to the video on youtube: youtube.com/watch?v=ftRZ...

- Claude Opus 4.5 on claude.ai (Pro plan): 12.3s
- Gemini 3 Pro on gemini.google.com (Pro plan): 14.3s

Claude models on claude.ai take a hit from an annoying bug where they reload the page after the initial load...

- Google search, featured snippet: 1.5s
- the default Mistral model on chat.mistral.ai, logged out: 4.4s. Congratulations to the Mistral team!
- Google search + stack overflow: 5.2s
- Gemini 3 Fast on gemini.google.com (Pro plan): 6s
- Claude Haiku 4.5 on claude.ai (Pro plan): 10.4s

The latency of LLMs is annoying when you use them often.

I tested a few LLMs, plus google search, on a simple question: "how do i reverse a string in python?". Results, from fastest to slowest:

... make the whole net-positive, though).
For instance, here's a modest one I generated after learning about the Shapley value. I had this diagram in mind and I wanted it to exist, so I could add it to my Anki card. Gemini 3 Pro created it for me

I've long believed that AI image generation has much bigger societal costs (loss of trust in the image media, very effective scamming) than benefits (AI images in blog posts, indie movies...?). I admit I didn't anticipate the generation of diagrams, which is a nice benefit (not nice enough to...