NeuralNate
16 subscribers
136 photos
20 videos
1 file
99 links
Download Telegram
GLM 4.7 is a beast, it is an open source model that is almost as good as opus 4.5 (10x cheaper) and better than sonnet 4.5 and is free on opencode for the moment.

TBH, who cares about intelligence advance, cost decrease is where the amazing stuff is at.
It's funny how sometimes it can make you feel your falling behind in both directions. Like the worst of both worlds.

https://x.com/i/status/2004607146781278521
😁2
NeuralNate
Photo
wait, it works exactly like death note

You send a label to Ryuk’s notebook (example from the README.md: label=something), and at the end of a Testcontainers test suite, when containers/networks/volumes/images are not needed anymore, Ryuk removes all these resources carrying the labels written in its notebook (the Death Note).
best feature in opencode

let's say you were being a bit of a dumbass and you were in a doom loop of 20 messages and code changes for an hour on something that could have been fixed with one line change.

you just go back in the timeline, undo all the code changes and chats you had with it upto that point you choose and act like you still have a sliver your pride left.
NeuralNate
Photo
192$ estimated from token usage, i just use the codex cli that comes with my siblings 20$ chatgpt subscription
From a 2 months 20$ subscription, I got almost 200$ of API credit usage.

If anyone around you has a chatgpt, Claude, Gemini subscription and they aren't using their respective cli's credits, you could just jump in for free.

People already buy and use it for the pro features on the websites, why let the API usage be wasted.

Even if you bought the 20$ plans, it is still a good investment.
NeuralNate
From a 2 months 20$ subscription, I got almost 200$ of API credit usage. If anyone around you has a chatgpt, Claude, Gemini subscription and they aren't using their respective cli's credits, you could just jump in for free. People already buy and use it…
Btw, the reason they're doing this is to promote their own cli tools and compete with each other. (I psiphon all their cli's into opencode tho, hehehe)

They have a boat load ship load of money to spend on dumb things and this just happens to be one of them. This is just a drop in the bucket.

If AI is going to ruin the ram, SSD, GPU market for us, might as well fight back from here.
NeuralNate
karma came just in time
for students, or former students with an institutional email, you can get the github pro subscription (10$) for free and it gives you 300 requests per month (a lot more generous that you think).

i have been using gpt 5.2 + minimax 2.1(free model on opencode/openrouter, better than 5.1 codex medium on benchmarks) and it is a really good combination.

github pro charges per request not by amount of tokens you use so for gpt 5.2 which is extremely token hungry, it is an amazing combination
Ever wondered which models do best on which languages, it's a bit outdated now but it shows the trends. 1
Just what I needed
🤣1
#tip

for large deterministic tasks, make the ai write a python script to do it instead of asking it do it one by one.

eg: instead of telling it rename large amounts of files, tell it to make a scipt to rename things based on something.

the script might crash your pc, but atleast it is deterministic, fast and won't bloat your ai's context.
2