NZ ☄️

16.1K posts

NZ ☄️ banner
NZ ☄️

NZ ☄️

@CodeByNZ

AI Insider / Tracking the race to AGI Startups • Agents • Open Source Sharing what matters before it goes mainstream

127.0.0.1 Katılım Ağustos 2024
778 Takip Edilen18.7K Takipçiler
NZ ☄️
NZ ☄️@CodeByNZ·
👀 OpenAI has made several adjustments to GPT-5.6 Sol usage. In a note to Codex and ChatGPT Work users, the company said it has rolled out inference optimizations that should deliver roughly 10% more usage across subscriptions. They’ve also reverted the context window for GPT-5.6 Sol back to 272k tokens (from the previously increased 372k), after noticing it was causing higher-than-intended usage. Additionally, OpenAI has adjusted reasoning effort settings and made fixes to multi-agent usage and auto-review to improve efficiency. The temporary 5-hour limit on usage has also been kept in place for now. These changes should meaningfully reduce the rate at which usage is consumed.
NZ ☄️ tweet media
Tibo@thsottiaux

Updates for Codex and ChatGPT Work users. No nerfing, only good stuff! - We have landed inference optimizations and are passing down savings to all the subscriptions for GPT-5.6 Sol. That should result in around 10% more usage on its own. - We noticed that by changing the context size limit in the product to 372k for GPT-5.6 Sol, up from 272k for GPT-5.5, it resulted in more usage being charged than intended. We have reverted to 272k and will work to roll back out to 372k in the days to come. You should notice that usage drains significantly less after this change. - To understand where the extra usage was coming from, we ran some experiments where reasoning efforts were changed (referred to as juice values under the hood) and have reverted this. - There is slightly more usage of multi-agent than intended in high and xhigh reasoning effort, we are fixing this going forward. Also fixing a small other thing we noticed with auto-review where we can be more efficient. And we continue to have the 5h limit temporarily not apply. Enjoy the rest of the weekend!

English
2
1
4
330
NZ ☄️
NZ ☄️@CodeByNZ·
@notjazii posted this same thing last week. people said i was hyping them.
English
2
0
2
42
NZ ☄️
NZ ☄️@CodeByNZ·
Rate your AI dependency out of 10 right now?
English
19
0
14
683
NZ ☄️
NZ ☄️@CodeByNZ·
What’s one thing you can do that no AI can do yet?
English
11
0
8
371
NZ ☄️
NZ ☄️@CodeByNZ·
been busy all day making the best out of gpt 5.6 sol. damn i love this model, worth every dollar.😄 what did you work on today ?
NZ ☄️ tweet media
English
2
0
10
392
NZ ☄️
NZ ☄️@CodeByNZ·
@TimJayas Claude for deep work yes i agree . But gemini, i don’t even use it anymore totally forgot i have the app 😅
English
0
0
0
25
Tim Jayas
Tim Jayas@TimJayas·
@CodeByNZ Claude Fabe for important work and Gemini for quick answers
English
1
0
1
32
NZ ☄️
NZ ☄️@CodeByNZ·
Which AI model do you actually trust with important work and which one do you just use for quick answers?
English
12
1
14
622
NZ ☄️
NZ ☄️@CodeByNZ·
@AskSnehasish Yeah, I trust them to help me move faster, not to be right every single time.
English
1
0
0
35
Snehasish Nayak
Snehasish Nayak@AskSnehasish·
@CodeByNZ Apparently, I just don't trust any AI model for the final output until I actually validate it.
English
1
0
1
31
NZ ☄️
NZ ☄️@CodeByNZ·
@juiceboy_of_abj I like that setup. No point burning your best model on simple questions.
English
0
0
0
31
Elijah 🌊
Elijah 🌊@juiceboy_of_abj·
@CodeByNZ I always use opus 4.8 for all my task While fable 5 is available I use it for complex task like payment int, and ask it to review a few logics I use sonnet for quick answers outside my IDE
English
1
0
2
82
NZ ☄️
NZ ☄️@CodeByNZ·
Any day I skip scrolling on this platform, a thousand and one things happen in the tech niche. WTF do you mean Claude extended Fable 5 access? At this point, they should just keep it.
NZ ☄️ tweet media
English
3
0
6
535
NZ ☄️
NZ ☄️@CodeByNZ·
@proxy_vector Makes sense. If you don’t measure it, it’s hard to know whether the model changed or your expectations did.
English
0
0
0
12
Rohan
Rohan@proxy_vector·
@CodeByNZ The giveaway is usually a regression in a specific task, not a vague “feels worse.” Keep a small versioned eval set for your real workflows; otherwise provider updates turn into anecdotes you can’t reproduce.
English
1
0
1
29
NZ ☄️
NZ ☄️@CodeByNZ·
Which model has gotten worse with updates and you noticed immediately but said nothing?
English
12
0
14
1.9K
Deepak K
Deepak K@deepakThamizhK·
@CodeByNZ claude around early 2025 got noticeably more cautious overnight and i just silently switched workflows lol
English
1
0
0
27
NZ ☄️
NZ ☄️@CodeByNZ·
@RanHasNotes I haven’t seen Kimi come up as much, but you’re definitely not the first to mention it.
English
1
0
1
21
NZ ☄️
NZ ☄️@CodeByNZ·
@AOluwaseun37257 Interesting. I’ve seen a few people mention hallucinations more often lately.
English
0
0
0
14
Allan Guerriero
Allan Guerriero@CCIE_24825·
@CodeByNZ github copilot. it has the memory context of a blacked out drunkard.
English
1
0
1
52
NZ ☄️
NZ ☄️@CodeByNZ·
@0xerik I don’t know about every model, but I’ve definitely had moments where an update felt like a step backward.
English
0
0
0
33
erik
erik@0xerik·
@CodeByNZ Literally every single one. Crazy that these companies can just make the models worse and not even tell us about it.
English
1
0
1
59
NZ ☄️
NZ ☄️@CodeByNZ·
@mretsal It definitely sparked a lot of debate when 4.7 came out.
English
0
0
1
21
Marcelo Retana
Marcelo Retana@mretsal·
@CodeByNZ Opus. When they went from 4.6 to 4.7. I immediately knew 4.7 was awful. It was a downgrade.
English
1
0
1
39
Paulake
Paulake@jetexpert7·
@CodeByNZ Deepseek it has been fucking terrible
English
1
0
1
40