18 • SWE • browser use guy intern @asciidotdev (YC F26) betterwright.com wanna have a call with me? cal.com/navster/1-hour-meeti…
- Tweets 1,292
- Following 289
- Followers 410
- Likes 147
Nav retweeted
Space Bunny strongly points to MiniMax 🐇
27/27 text probes matched M3/M2.7 tokenization. Image-token counts closely matched M3, plus the exact same image-decoding error (2013).
Likely MiniMax-family. Exact model and ownership aren't officially confirmed.
5
7
76
6,173
Gonna be helping out with something soon! And my condolences for all the different agent companies 😔
2
9
455
this dude just put a 300B param model on a single rtx 6000 PRO and 128 gigs of ram and got 40tps?? hes cracked af @0xSero
I shoved MiMo v2.6 Flash UNQUANTIZED onto ONE RTX pro 6000 and 128gb RAM and after a few days of nonstop work got 40 tok/s decode speed and 2000 tok/s prefill with 512k context window (that is perfectly usable to me!!)
Fully open source and reproducible, give it a try! github.com/SSHdotCodes/mimo-…
4
374
for those who think astra is cheaper than fable for coding.. no its not... fable 5.1's super cheap caching really brings it ahead despite astras token efficiency
1
3
366
Browser use that is up to 90% MORE token efficient and FAST.. AND its FREE and OpenSource??
Real time demo using Qwen 3.8 27b, hosted locally.
6
2
25
5,045
Site: betterwright.com
github: github.com/BetterWright/bett…
sponsors: github.com/sponsors/BetterWr…
3
301
site: betterwright.com
github: github.com/BetterWright/bett…
sponsors: github.com/sponsors/BetterWr…
26
Nav retweeted
肝了一夜！Dario大战Altman！
Seedance2.5的参考视频没法去掉音效，所以有点瑕疵
无论如何，Claude封了我两个号，我选Altman赢！
903
1,137
15,724
6,094,556
Nav retweeted
yeah officer so he was using astra, and i told him he should've used fable, because fable writes more mergeable code, you know how gpt models love their tests? i also saw him on the terminal, we have guis now, officer have you heard about t3ode? it's the all in one interface for using your codex and claude subscription
88
69
3,067
70,061
u can try it at github.com/BetterWright/bett…
15
had someone claim jev beat betterwright in browser use by hardcoding jev to do a task.. ai psychosis is real..
1
4
612
Are u someone who wants to give your product/agent better browser use or know a founder who could benefit from better browser use. Please ping them in the comments here! Ill help them for FREE
1
463
what do you mean i can get 4 vcores and 8 gigs of ram for under 4 cents an HOUR???
boat.dev
3
19
2,845
Browser use is now 40% faster!
We've made changes to BetterWright to make our browseruse even faster!
github.com/BetterWright/bett…
3
31
5,947
People are so jev pilled.
JEV WILL NOT REVOLUTIONIZE COMPUTER OR BROWSER USE.
Is it good for simple stuff?
Yes absolutely and does those fast infact but computer use and browser use are domains that require lots of context and multi step planning and outputs that jev simply cannot do..
Why dont you pair it with an LLM then??
I did try.. i just find that for complex tasks ur better off just having the LLM do everything rather than orchestrate jev. And i tested it with the already optimized BetterWright and the results were worse or BARELY better for SOME tasks..
2
6
536