Good experience working alongside Proximal folks in uncovering what the current frontier looks like
We are releasing FrontierSWE v2, our updated ultra-long horizon coding benchmark

V2 features an expanded task suite and improved methodology. We see large performance gaps between frontier models, with Claude Fable 5.1 leading by a wide margin
0
0
1
97
Pretty cool that Instinct gives you a persistent sandbox. They’re using E2B deployed on GCP.
It comes with a lot of bells and whistles I cannot imagine this being a service cheaper than $100/month.
1
0
0
50
Replying to @BillyBakht
Quick update: it only stays persistent for a few minutes
0
0
0
14
Ox Alpha's reasoning traces are extremely coherent.
Interesting to see that they're not worried about distillation either, maybe going to be an open-weight release?
0
0
0
58
Why has replit not bought a tld yet
0
0
0
44
wonder if Claude’s watermarking makes it more deterministic
0
0
0
41
I wonder if there will companies offering undoing Claude's watermark and it will basically just be a cat and mouse game
0
0
0
23
it is becoming increasingly clear every passing day that Opus 5 sucks a lot
0
0
2
32
deepseek-v4-flash was already lowkey favorite model. excited to try out the new version.
0
0
0
95
What do you call it when someone tries to drown out the issue at hand with unnecessary and tangential details just to do things their way?
Because that is what it feels like talking to Claude lately.
0
0
0
41
claude's been drinking
0
0
0
22
fable just flexing some chinese
0
0
0
47
anytime I open the claude web>settings>usage page, my computer slows down so much.
0
0
0
46
Super fascinated by how fast AI Overviews are on google. Especially the "People also ask" section. Genius level caching and engineering happening there.
0
0
0
44
Bilal Bakht Ahmad retweeted
They didn’t mean pause AI research, they meant pause *your* AI research
48
329
4,931
111,705
Imagine if releasing Fable/Mythos makes more people realize they don't need the best model most of the time.

That would not be a good outcome for western frontier labs.
0
0
0
37
opus 4.8 explanations are unreasonably dense
0
0
0
31
The most meaningful thing since Opus 4.5 in november has been codex remote connections.

Claude's remote control has nothing on it.
0
0
0
39
what scares me is that the behavior one might build a reliance on of closed-weight AI models might not exist tomorrow.
0
0
0
22