Get live statistics and analysis of Neko Legends's profile on X / Twitter

For finding bugs, Roo Code + GLM 4.6 thinking beats both Minimax m2 AND the latest Kimi K2 Thinking. I have stopped using Kilo Code now after this, although it is suppose to be a fork of Roo which is a fork of Cline; there is some issue with tool calling on Kilo Code (windows). Kimi K2 Thinking fix attempt (fail): Cost $0.13 , time: 1 min 38 sec GLM 4.6 fix code: (success with try again prompt) Cost $0.23 , time: 1 min 6 sec But of course that's just what it would cost if you pay from official sites. These models are INCLUDED with the subscription. I have scoured far and wide for the best AI deals and @NanoGPTcom is still the best. 8$ a month for 2000 uses a day is insane (can be unlocked to 60,000 a day but that's the monthly limit); some competitors are like 2000 a week and some even 20 a week (*cough new Codex usage changes). If a single bug fix averages 15 cents, and you max it out at 2k a day, that's like 300$ a day saved, and if you only use it like a normal dev and only do 200 a day that's still like 30$ a day savings. I only saved 10$ today because I was mostly using kimi to try and fix stuff but it kept failing. It's great for refactoring large code to make them more modular, but GLM 4.6 is still king in my eyes for bug fixing... for now! I asked ChatGPT Codex like 5 questions and it's already out of commission for a week with 16% usage left AND doesn't even fix my shit, yet cost 20$ a month. I'm 2 weeks into using this thing and I feel like I fell for the trap; I'm addicted to using it as if I'm actually making money the more I use it and there is no fear of asking the wrong questions, which may cause you to end up using a few bucks for a single FAILED query on pay per use plans. This 8$ allows you to figure out which model works best for you, and new models are added FAST. I can't stress how much of a steal it is and if I was a competitor and wanna do evil I would sign up for a ton of 8$ accounts and try to break the company by maxing/spamming/abusing the limits. @grok I'm sure companies are already doing this in the AI space, you know, like API abuse/sabotage; do you know of any instances of this that people has talked about? And lastly, shameless shill of my referral code for anyone to use to get 5% forever discount on paid only models. (I don't think it applies to the 8$ subscription fee though). I wished someone gave me 5% forever discount to use other paid models! nano-gpt.com/invite/KaWMKneR

My initial thoughts on the new kimi-k2-thinking - TLDR version: It's good and responded with a working solution to a problem I asked, unlike other models, but probably need to use turbo mode for speed. Long version: I asked it to fix a bug (from my two sentences of input), it thought and worked for 20 minutes, cost me 30 cents, and came up with a workaround solution (not ideal fix but works?) for a fix in a file of about 500 lines of code. It at least came with a workaround while other models such as chatgpt codex, minimax m2 and older kimi failed to fix it at all - they al lied that they fixed it but in reality code changes were made but the behavior was the same. @Kimi_Moonshot , good job on this model! My Christmas wish is to have this model included in the @NanoGPTcom subscription. Maybe up the pricing to 10$? 😅

For finding bugs, Roo Code + GLM 4.6 thinking beats both Minimax m2 AND the latest Kimi K2 Thinking. I have stopped using Kilo Code now after this, although it is suppose to be a fork of Roo which is a fork of Cline; there is some issue with tool calling on Kilo Code (windows). Kimi K2 Thinking fix attempt (fail): Cost $0.13 , time: 1 min 38 sec GLM 4.6 fix code: (success with try again prompt) Cost $0.23 , time: 1 min 6 sec But of course that's just what it would cost if you pay from official sites. These models are INCLUDED with the subscription. I have scoured far and wide for the best AI deals and @NanoGPTcom is still the best. 8$ a month for 2000 uses a day is insane (can be unlocked to 60,000 a day but that's the monthly limit); some competitors are like 2000 a week and some even 20 a week (*cough new Codex usage changes). If a single bug fix averages 15 cents, and you max it out at 2k a day, that's like 300$ a day saved, and if you only use it like a normal dev and only do 200 a day that's still like 30$ a day savings. I only saved 10$ today because I was mostly using kimi to try and fix stuff but it kept failing. It's great for refactoring large code to make them more modular, but GLM 4.6 is still king in my eyes for bug fixing... for now! I asked ChatGPT Codex like 5 questions and it's already out of commission for a week with 16% usage left AND doesn't even fix my shit, yet cost 20$ a month. I'm 2 weeks into using this thing and I feel like I fell for the trap; I'm addicted to using it as if I'm actually making money the more I use it and there is no fear of asking the wrong questions, which may cause you to end up using a few bucks for a single FAILED query on pay per use plans. This 8$ allows you to figure out which model works best for you, and new models are added FAST. I can't stress how much of a steal it is and if I was a competitor and wanna do evil I would sign up for a ton of 8$ accounts and try to break the company by maxing/spamming/abusing the limits. @grok I'm sure companies are already doing this in the AI space, you know, like API abuse/sabotage; do you know of any instances of this that people has talked about? And lastly, shameless shill of my referral code for anyone to use to get 5% forever discount on paid only models. (I don't think it applies to the 8$ subscription fee though). I wished someone gave me 5% forever discount to use other paid models! nano-gpt.com/invite/KaWMKneR

My initial thoughts on the new kimi-k2-thinking - TLDR version: It's good and responded with a working solution to a problem I asked, unlike other models, but probably need to use turbo mode for speed. Long version: I asked it to fix a bug (from my two sentences of input), it thought and worked for 20 minutes, cost me 30 cents, and came up with a workaround solution (not ideal fix but works?) for a fix in a file of about 500 lines of code. It at least came with a workaround while other models such as chatgpt codex, minimax m2 and older kimi failed to fix it at all - they al lied that they fixed it but in reality code changes were made but the behavior was the same. @Kimi_Moonshot , good job on this model! My Christmas wish is to have this model included in the @NanoGPTcom subscription. Maybe up the pricing to 10$? 😅