quite good! Sol & Astra were substantially cheaper and more token efficient than Claude benchmark runs
6.2 Sol: ~$6, 29.5M tokens in (28.9M cached)
6.2 Astra: ~$13, 12.3M tokens in (11.8M cached)
5.5 Opus: ~$62, 128M tokens in (127.9M cached)
5.2 Fable: ~$117, 167M tokens in (167.9M cached)
GPT-6.2 is a massive jump for OpenAI in my personal benchmark: recreating a Nord Stage 4 in the browser.
Here are the latest results, with Astra 6.2 at #2 and Fable 5.2 at #1.
I've been running the same benchmark on every new major coding model: recreating a Nord Stage 4 🎹 in the browser from one detailed photo, a 76-page manual, and several JSON specs.
Here's the latest generation. Guess who takes the crown 👑
@0xZoZoZo you mean these calibrations? fully designed by claude. I said to make it sound like real physically modeled instruments and compare to real recordings. I still had to steer several times when it sounded too robotic or the violin bowing was wrong, etc
https://t.co/BzjgGwM9lf
1/ I asked Fable 5.2 to write me a symphony. And to build an orchestra where every sound is synthesized live from physics, with a full 3D WebGL view of each instrument. This is Toward Morning.
@vitorialima_42 ~half done with opus 5.5 last weekend before fable 5.2 came out!
I rarely use sol 6.1 or astra 6 anymore (except to get a second opinion), since I think opus 5.5 is just better overall, especially for creative projects. fable 5.2 is a nice upgrade but pricey for most things
Faster Notion pages! ⚡️
Previously, we rendered every block on a page at once, which could dramatically slow things down on long pages.
Now we render as you scroll, reducing memory usage and making the whole application feel snappier.
Give it a try and let us know what you think!
@helenmag@claudeai@NotionHQ really neat, thanks for sharing!
I'm sure you're gonna have some fun updating this w the Claude Duet launch :-)
do you mostly talk to Floofy directly?