Downloads of top open LLMs on HuggingFace drop ~30% — counting change or real decline?
TheZachMueller · x · 2026-10-11
Nat Lambert reports a 30% decline in HuggingFace downloads across top open LLMs, continuing over time, possibly a counting change. TheZachMueller hypothesizes: fewer architecture experiments plus ever-larger models mean only big providers and hyperscalers download the largest models, reshaping the download mix.
More from Infra
- Codebase test: Telnyx-hosted GLM runs 9% cheaper, 5% faster than OpenAI — SucceededMind · 2026-10-11
- Nvidia said to go head-on with frontier labs as labs build their own ASICs — pzakin · 2026-10-11
- Local LLM user on RTX 5090 weighs Qwen 27B vs Flash Next for coding: is a bigger model worth it? — MasterNomie · 2026-10-11
- AirLLM runs 70B models on a 4GB GPU via layer-wise inference, scaling to 405B on 8GB — JensHonack · 2026-10-11
- How close can a 5090 + 32GB RAM rig get to Codex? Dev seeks local agent coding setup — GooseEggs712 · 2026-10-11
- Chart: how compute shifted from pretraining to post-training over two years — japie06 · 2026-10-11