HACKER Q&A
📣 throwaw12

How do you find SOTA LLMs for a task?


There are thousands of models at the moment available at Hugging Face. But whenever I need a model for specific task, I am struggling to find SOTA model, can you recommend me how to find it?

I am not ML practitioner, I just need models for my work, for example for coding, I know we can use Claude/Gemini models, but sometimes I want to compare them to SOTA open source, every week something better is coming and reading articles from month ago or finding LLM leaderboard for a specific task is difficult sometimes. I think some kind of model picker already exists, but don't know where


  👤 Oras Accepted Answer ✓
I usually go to OpenRouter usage to learn that by category https://openrouter.ai/rankings

Scroll down to categories, and select from the dropdown on top right of the chart.


👤 incomingpain
For open source, you're not going to see stats well online. Openhands + devstral doesnt touch the internet, so wont make it to many stats.

You can look at benchmarks.

https://livebench.ai/#/?Agentic+Coding=a

Keep scrolling until you see something your size. Deepseek R1 is nice, but 600B isnt running on my hardware. You'll also notice they arent doing everything. dominated by the Saas options.

https://huggingface.co/models

This is sorted by trending by default. This tends to help show interest but not necessarily the best.