Nigmat

too ez for gpt in codex now. toooooo ez

@thegenioo definitely 5.5. I'm not a billionare, no chance to use opus 4.7

Is it GPT-5.5 vs Opus-4.7 or GPT-5.5 Pro vs Opus-4.7 ??

@abeni_t @thsottiaux Codex with 5.5 is strong as f, I write agentic skills in my researching area (like bioinformatics)， With that the dominant agent can keep watching the whole process and record every mistakes and what they actually do. When I wake up I can see the new article and new idea

@OmniScopeBio @thsottiaux How you run it while sleeping

Looking at the traffic dashboard for Codex just now, it would be scary if we didn't have a lot more compute coming online in the coming weeks. All according to plan fortunately.

@sama Absolutely, especially Agent friendly is the main topic in the near future, compute should not be wasted in useless GUI design just for entertainment. It's not effective

feels like a good time to seriously rethink how operating systems and user interfaces are designed (also the internet; there should be a protocol that is equally usable by people and agents)

@iruletheworldmo It's too cheap that I have no reason to try others

this is a great read, the image is pretty cool too. deepseek is a serious ‘threat’, or competitor if you will. make sure to bookmark. it’s not quite sota but it’s so cheap it may not matter.

Doing a research that intergrate MOE in Oncology, hope there will be a good outcome

@bridgemindai They do most of the efforts in KV cache, ability might increase in the near few months with stronger post training

DeepSeek V4 Pro debuted at #14 on LMArena Code. Remember when DeepSeek was supposed to be the future of open source AI? The model that was going to compete with frontier labs? Ranked below a $0.33 input model from Alibaba. DeepSeek fell off hard.

@victor207755822 Definitey, AGI belongs to everyone, not like the anthropic. Ego company acts like know all the things

DeepSeek-V3: Dec 26, 2024 DeepSeek-V4: Apr 24, 2026 484 days later, we humbly share our labor of love. As always, we stay true to long-termism and open source for all. AGI belongs to everyone. ❤️🌍 #DeepSeekV4 #AGIforEveryone #OpenSource

@chatgpt21 Yes, We got v4

1. What?

@Brooooook_lyn Yes. I tried it in the web maybe 2 weeks ago. It's freaking fast

DeepSeek v4 in API? So fucking fast 🤯 Ascend or Nvidia?

@iruletheworldmo @sama finally, I hope u can get back to ur normal nose tdy

@sama i’m surprised it could get this much better. makes you think what future intelligence can unlock. helluva week sammy. happy birthday btw.

Images 2.0 really got over some important qualitative threshold for me that I didn't know existed.

@sama just do it!!!!

we love seeing our users win. we want to give you the best tools, lots of compute, and watch you do the magic.

Show us more SAMA, just smash all of us

@blueshades2020 @chadptg @chatgpt21 Maybe, Just waiting for release. Let's see, I hope it's sota

@chadptg @OmniScopeBio @chatgpt21 it's their internal codex that will be totally different than how we use and also its models will be different. internally it seems they are still using 5.4 xhigh probably because Spud is not really good at coding but good at emotional intelligence and creativity simila to 4o

Full porting spud into codex

@nicdunz Elon says grok 5 will be agi, so now just call it agi 0.43

if they dont call tomorrow agi 0.5 i will be disappointed

@chadptg @chatgpt21 BTW, was Heavy Thinking (Web) already a thing before? I just noticed it, but I’d never used it. Is it new, or did I somehow miss it earlier?

@OmniScopeBio @chatgpt21 yeah i think we’re just joking haha

@chadptg @chatgpt21 yeah for sure🥰

Uhhhh, I don’t remember having access to Thinking Heavy. Maybe I just haven’t used Thinking on web in a while. Can someone tell me when they rolled it out?

@chadptg @chatgpt21 Chill... just kidding. What I’m saying is, they’re still not smart enough to port Spud into Codex using Spud from the cloud. Like, you shouldn’t need to import Spud into Codex if you can just rely on the ability of spud

@OmniScopeBio @chatgpt21 now tell me how they would use 5.5 to port itself