44za12

Where are the benchmarks? Does it just speed up inference and reduce token usage or does it seem to improve accuracy on certain tasks as well?

Built rightsize exactly for this. https://nehmeailabs.com/right-size

Others have already pointed out the absurdity, but just to put it across. Why not just ask first agent to update AGENTS.md adding context and next steps, and ask the next agent to continue from there?

+1 came here to say this, I opened the link expecting some technical breakthrough. Misleading click bait title.

Location: UAE Remote: Preferred Want to relocate: No

Philosophy: Brutally Efficient

More: https://aazar.me

I read it as an article in defence of boring tech with a fancier/clickbaity title.

Here’s the more honest one i wrote a while back:

https://aazar.me/posts/in-defense-of-boring-technology

Specialised models easily beat SOTA, case in point: https://nehmeailabs.com/flashcheck

All of us use the same keyboards more or less, maybe us randomly typing a large number is not as random as we would like to think. Just like how “asdf”, “xcyb” are common strings because these keys are together, there has to be some pattern here as well.

Yes, I included a 'Model Selection Cheat Sheet' in the README (scroll down a bit).

I map them by task type:

Tiny (<3B): Gemma 3 1B (could try 4B as well), Phi-4-mini (Good for classification). Small (8B-17B): Qwen 3 8B, Llama 4 Scout (Good for RAG/Extraction). Frontier: GPT-5, Llama 4 Maverick, GLM, Kimi

Is that what you meant?

This is the way. I actually mapped out the decision tree for this exact process and more here:

https://github.com/NehmeAILabs/llm-sanity-checks

For simple extraction tasks, a delimiter-separated string uses 11 tokens vs 35 for JSON. Output tokens are the latency bottleneck.

Love the minimalism.

Shameless plug.

I’ve been using a cli tool i had created for over 2 years now, it just works. I had more ideas but never got to incorporate those.

https://github.com/44za12/horcrux

Have been using remove.bg for this for years now.

Like a sempahore?

I’ve had great luck with all gemma 3 variants, on certain tasks it the 27B quantized version has worked as well as 2.5 flash. Can’t wait to get my hands dirty with this one.

Can you benchmark Kimi K2 and GLM 4.5 as well? Would be interesting to see where they land.

That was quick, vibe coded, I presume?

The ability to submit a story using a curl would have been fun.

Tried that it’s taking exactly as much time as my program.

I have breach parser that i had written to parse through over 3 billion rows of compressed data (by parsing i simply mean searching for a particular substring), I’ve tried multiple LLMs to make it faster (currently it does so in <45 seconds on an M3 pro mac) none have been able to do that yet.

https://github.com/44za12/breach-parse-rs

Feel free to drop ideas if any.

Why not use something like litellm?

Making a model better than proprietary models is in fact making a model better than their closed source models if you believe the benchmarks.

With claude code and Kilo as well. I’m using moonshot’s API.

I am on the pro plan, I was considering Max, but then i found kimi and I’m getting used to it.

I pay for Cursor, OpenAI and kimi (to use with Claude Code), OpenAI is good with quickly refining my thoughts, Cursor’s subscription I’m reconsidering to cancel bought it for Claude but the rate limits are making it impossible for me to find it useful. Kimi is what truly surprises me, Claude code shows this conversation costed you $500 (based on Opus usage which is mapped to kimi k2) while I’ve barely spent $2. I have Ollama as well, majorly to quickly test small models that could be improved for our usecase through finetuning.

Is it something like tree of thought reasoning?

I don’t get the hype with OpenAI OSS, they would never make a model better than their proprietary models open source, and the other open source models beat GPT and family so why the wait?

“I used all the top tier ‘Open Source LLMs’ and they suck I was right, like always, LLMs suck so hard, my job is safe!”

Honestly, the fact that “scan everyone’s encrypted messages” keeps coming back like a bad sequel says a lot about how governments see privacy, as a speed bump, not a right. Every time this proposal pops up, it’s wrapped in “think of the children” language, but the end result is always the same: less security for everyone, more power for people who already have too much. At some point, you have to wonder if the real goal is to make everyone so paranoid they just stop communicating altogether. Here’s hoping enough people still care to push back—otherwise, we’ll all be talking in code again, and not the fun kind.