← back

WTF do people use Open Models for??

2.4K views · Feb 22, 2025 · 28:00 min · Watch on YouTube ↗
Takeaway

Open models in production are mostly running creative writing, roleplay, and companionship — and enterprises pin year-old models like Mistral Nemo for stability.

Summary

  • Eugene (Featherless AI) shares data on actual open-model usage: 50K+ models uploaded to HuggingFace per month; DeepSeek R1 alone moved 2.74 exabytes via 4M downloads of a 685GB model.
  • For individuals (flat-rate $25/mo, no per-token pricing), DeepSeek R1 dominates, then Llama 3, Mistral Nemo, Qwen — choice driven by vibes/fame not MMLU or cost.
  • Commercial usage is dominated by Mistral Nemo (Apache 2.0, pre-Llama-license-restriction) — Enterprises pin model versions for years for stability ('don't break it') and Llama Guard 2 stays for safeguard tutorials.
  • Top non-coding workloads (30-40% of all traffic): AI roleplay/companionship/therapy/creative writing — 60%+ users are women, mirrors romance-novel market, not the stereotype.
  • Use cases beyond coding: comfy UI image gen, agents/work, journaling/therapy — 'no one cares about MMLU, this segment is all about Vibes'.
open-modelshuggingfacecompanionship-ai
Original description
from Eugene Cheah, CEO, Featherless.ai

Covering the open weights landscape. What’s being downloaded. What’s being used.

The heritage and what’s to come next