Weekly Discussion Hub - Model Performance & API Services - Starting July 27, 2025

[scarlettturner]

Welcome to our weekly community discussion for model performance and API service reviews.

All general discussions about models and APIs should go here. Individual posts asking “Which model works best?” will be removed from the main forum.

Note: This thread is not for promoting your own services. We might allow announcements for genuine new services occasionally, but don’t expect advertising posts to stay up.

Thread Organization Guide

Find the category that matches your topic in the comments below:

• LARGE MODELS: 70B+ – Talk about models with 70 billion parameters and above
• MEDIUM-LARGE: 32B-70B – Discussion for models between 32B and 70B parameters
• MEDIUM: 16B-32B – Models in the 16B to 32B parameter category
• SMALL-MEDIUM: 8B-16B – Models ranging from 8B to 16B parameters
• SMALL MODELS: Under 8B – Discussion of compact models below 8B parameters
• API SERVICES – Share experiences with API providers including costs, speed, and availability
• OTHER TOPICS – Everything else about models and APIs that doesn’t match above categories

Post your questions, reviews, or suggestions under the right category comment below. This organization helps everyone find relevant information quickly.

Let’s get started!

[lucasg]

Thanks for bringing back the weekly hub. The parameter-based categories work way better than that subject sorting we tried a few months ago. Much easier to compare similar-sized models without all the random off-topic stuff cluttering things up. One quick suggestion - maybe add a note about where mixture-of-experts models go? People keep posting Mixtral in different spots depending on if they’re counting total vs active parameters. But yeah, this setup’s definitely better than all those scattered individual posts we used to get.

[Luke_Brilliant]

finally! i was getting sick of scrolling through dozens of “what’s the best coding model” posts every day lol. the parameter grouping makes sense, but shouldn’t we mention context length too? a 7b model with 128k context hits way different than 32k for most use cases. good to see this organized again though.