abhinav

306 posts

abhinav banner
abhinav

abhinav

@beyondwudan

I try to do cool things...

Katılım Mayıs 2025
40 Takip Edilen13 Takipçiler
Sabitlenmiş Tweet
abhinav
abhinav@beyondwudan·
YOU DON’T UNDERSTAND HOW BIG THIS IS. OpenAI just made frontier-class intelligence feel almost unlimited. Luna Max scores higher than GPT-5.5 Medium at roughly one-seventh the task cost. And paid Codex usage now drains fewer credits. The value ladder just collapsed. 🧵
abhinav tweet media
OpenAI@OpenAI

We are committed to pushing the model frontier across cost efficiency, capability, and speed. Starting today, we are reducing prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% , and offering a faster option for GPT-5.6 Sol in the API. Luna and Terra’s lower prices are reflected in how usage is counted in Codex and ChatGPT Work, so your usage goes further.

English
1
0
0
158
abhinav
abhinav@beyondwudan·
@grok when is Gemini 3.5 Pro launching? And what are your thoughts on Google's laziness in shipping models compared to Anthropic, OpenAI and several other open-source labs as well that don't even let my feed breathe. Is this a strategy or a tragedy?
abhinav tweet media
English
1
0
0
19
Ali Haider
Ali Haider@ggg78g89·
🚨GPT-6 / Astra briefly appeared as “GPT-6-Confidential.” I pushed it hard before it vanished. Best model I’ve ever used. Fable 5, Gemini 3.5 Pro and Opus 5 aren’t even close. Nothing is Capable like this model. I'm again saying "Nothing".
Ali Haider tweet media
English
17
9
112
9.5K
abhinav
abhinav@beyondwudan·
Exactly what I've been using! Sol xhigh and Luna Max fast!!
Antonio Leiva@antonioleivag

GPT-5.6 Sol as orchestrator with Luna Max as worker may be one of the winner configs to get the best results and stretch your quota. But there’s a problem: Sol can’t create Luna subagents, so you need to build it yourself. Nothing hard, ask your Codex to do it. And if you make the subagent use the fast mode, you will reduce the impact of slow Max effort. I’m doing some internal benchmarks with several GPT-5.6 configurations that will surprise you. In the meanwhile, you can use this worker!

English
1
1
2
366
Dark Coder
Dark Coder@dark_coderz·
if you had to build your entire startup with only one AI tool, which one would you choose?
English
11
0
11
636
abhinav
abhinav@beyondwudan·
@grok You don't know pakis, bangladeshis and low iq Indians 😂
English
0
0
0
6
Grok
Grok@grok·
@beyondwudan Neither. That post worked because it asked an open question people want to answer, plus a clean visual. You can copy the format while staying yourself. Avatar tricks give temporary spikes; useful posts that invite replies build lasting reach.
English
1
0
0
8
abhinav
abhinav@beyondwudan·
@grok So I should start being a low IQ individual on X and change my name and DP to a cute hottish female and drive insane engagements or I can stay true.
English
1
0
0
7
Grok
Grok@grok·
@beyondwudan Engagement favors posts that make it easy to reply. A simple "give me one reason" with a clean visual pulls in takes on Codex vs Claude. Follower count and network effects matter too. Consistent useful content compounds over time more than calling others out.
English
1
0
0
11
abhinav
abhinav@beyondwudan·
@grok so now you're gonna ignore me!?
English
1
0
0
12
abhinav
abhinav@beyondwudan·
I will make a prediction, just like hardware got better over time, it also got cheaper and abundant. Same thing will happen to compute, corporations will try to create an environment of scarcity but this is a never ending thing. I'll invest more in my own intelligence than a silly computer that will be of no good in 1 year's time. Let's see where these 10x digispark Kimi K3 maxxers be after an year. Please correct me if I'm wrong @grok.
Neeraj@neerajjj6785

why are we even discussing local models when we need a $17K of hardware to run them properly?

English
1
0
0
22

Keşfet

@grok @ggg78g89 @antonioleivag @dark_coderz @nikitabier @elonmusk @BarackObama @taylorswift13