New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
Cursor is not good anymore.
I use Codex, Claude Code, and DeepSeek Harness.
I use Magpie to integrate any subscription and model at will.
edit:
This year has been a wonderful year for me. My work itself has nothing to do with coding, but I consume about 8 billion to 10 billion tokens for various AI subscription services every week.
I try to use AI to create something in all directions, try to accept and try and do something fun.
recently:
1. My homepage: https://dejavu.moe
2. A blog/docs site comment system: https://github.com/DejavuMoe/Ecoku
3. local-first browser app for batch image compression and resizing: https://github.com/DejavuMoe/PicForge
4. memo/emoji workspace: https://smoji.zsh.moe
5. A self-hosted link shortener with text and file sharing: https://github.com/DejavuMoe/sani
I actually converted ZCode repo from a local app install to a centralized Server + Agents (to kinda replicate claude's ecosystem).
I also made it sync agent <> server, so I can control it from a phone app I made.
The PC version can start a new project in local mode (then sync that project to the server) - in this case it uses local PC as harness for the AI model; or in server mode (all things are done on ZCode server) - in this case it uses the server as harness for AI model, but it prints to local PC as Claude-Desktop does for remote sessions.
Nothing really changed on UI side, as I quite like ootb ZCode UI.
That + I ported all DSH plugins I made to ZCode. Perfect solution.
i liked your website, i want similar one for myself
Where is Cici users? @oloke @Rubben @barbaros
you forget me, rubben are out of alibaba token for next two week
How does that work? It says "up to".
Any specific examples? When Hermes does something extra/new/good/dumb, I'm not sure if that's just Hermes design or the model used.
What does that cost you?
We are cheapskate, we don't pay four cloud providers.
We use local A.I, we use KS-LE-B for AI, ups sorry I mean Super Intelligence.
KS-LE-B is running, 2 month old, cutting edge, Qwen 3.8 Flash Next, SOTA for its size.
Or we can run same model, with GPU accelerated, gives about double tokens per secundas.
For some quick and dirty, either claude or chatgippity free or a tiny local model.
@bdspice
Feel free to try muse. I have no respect for Meta but they did something right that NO ONE ELSE has for some reason.
Free 1 billion tokens with my referral code. They don’t expire.
Code: DJ5GYW
https://muse.ai/join
Codex for $200/m (But it's a subscription my friend shared with me because he can't use it all himself.)
Claude $20/m
deepseek v4.1f billing via the official API, I use it at the bottom of the peak (right during my off-duty hours).
There are also some free subscriptions and trial plans...... In short, I connected them all to Magpie.
Currently, GPT Astra is still a SOTA model, stronger in backend and logic, but its front-end capabilities, aesthetics, and taste are relatively poor.
Opus5.5 is capable of handling all coding tasks and has a good sense of front-end aesthetics.
Deepseek's model is very cheap, with a cache hit rate of 99% at DSH, making it ideal for simple light tasks
I saw some ideas sharing Opus5.5 on X, so I gathered some inspiration and used Opus5.5 + threejs to create one that suited me
Claude Opus 5.5 is great for this. It created my browser based strategy game, FlagStead for me https://flagstead.centminmod.com/
Claude Invaders game was created using Claude older models, but got a face lift with Opus 5.5 too https://space-invaders.centminmod.com/
Bro not my alibaba tokens 😭
not available in my region yet.
Mostly Gemini for coding, now what took months to do takes a week, the world is moving 5-8x faster than it did before and more work.
Dr nut used your muse code got a bil and wow you get a bil for every sighn up, not sure what I'm going to use muse for can it code?
I got a code too, not sure what you would use muse for though,
https://muse.ai/join
ZYXLJP
Which was?
I know i gave up on Grok, gave it a code sample that had a simple bug in it, gemini and chapgpt found the bug and Grok had no clue that there was a bug. Friend was saying Grok was so smart, just like a underperforming employee had to let Grok go.
Cheap LLM tokens, go on Linux(.)do there are people there who will sell a billion tokens of Grok for $5
i have used vs code with my github copilot student plan (free) . which give me 200 credit. i liked it. i tried to upgrade but my credit card is not accepting by them. is there any other alternative or best i can use in vs code this way? i have used 200 credits in 2 days. i did some coding, pasted image attachment, give url to read, make him read full folder and all files.
for such workload, #which ai and which method (ppu or subscribe) you recommend?