Daily download tracking since 28 Jul 2026

Which open models are actually winning — measured daily, not guessed.

We snapshot downloads, likes and file inventories for the top 1,040 Hugging Face models every night at 04:00 UTC. That history powers trend charts, VRAM estimates and side-by-side comparisons you can't get from the model card.

Catalog downloads · trailing 30d
1.8 B
▲ 8.3% vs previous 30d
last 44 days
26,287
compare pages
0
real outputs
44
days tracked
#Model30dΔ 7dTrend
01 NVIDIA-Nemotron-3-Nano-4B-BF16 nvidia · text-generation 2.8 M ▲ 636.8%
02 Llama-3.1-8B-Instruct-4bit mlx-community · text-generation 905 K ▲ 210.8%
03 lcm-lora-sdv1-5 latent-consistency · text-to-image 88 K ▲ 106.5%
04 Ornith-1.5-9B-NVFP4 ornith-ai · text-generation 476 K ▲ 106.4%
05 Qwen3.8-27B-iMatrix-NVFP4-MTP-GGUF cdiamond · image-text-to-text 2.1 M ▲ 106.0%
06 miaomiao-harem jd09876 · text-to-image 35 K ▲ 80.7%
07 Ornith-1.5-397B ornith-ai · text-generation 499 K ▲ 76.0%
08 glm-4-9b-chat-IMat-GGUF legraphista · text-generation 993 K ▲ 71.5%

Fresh comparisons

For machines
Read-only JSON, no key, 60 req/min. Snapshots also ship as a daily CSV dump.
Sponsored · GPU cloud
Rent an A100 by the hour
Run 70B models without buying hardware. From $1.29/hr.
Data from the public Hugging Face API, snapshotted daily. VRAM figures are estimates — see /methodology.