Hugging Face downloads, last 30 days / read 2026-10-03
Most downloaded on Hugging Face
Open-weight models ordered by Hugging Face's own download count, each joined to the smallest device in this catalogue that holds it. Downloads measure popularity, not quality or use.
Where the order comes from
The ordering is Hugging Face model API’s own thirty-day download count per repository, read on 2026-10-03 and cited on each model’s page. It is not this site’s traffic, and it is not a vote — a popularity list built from our own visitors would be unfalsifiable by you and mostly noise anyway.
What it is not: Repository downloads measure attention on one host, including mirrors, CI and evaluation runs. They are not a measure of quality, and a model distributed mainly elsewhere is undercounted.
Models with a published count3270 without one
Of the top 5046run on 24 GB or less
Sizing formatQ4_K_Mor the nearest published
Reference context8,192tokens, no offload
Downloads in the last 30 days
top 50| # | Model | Downloads / 30d | Format | Needs | Smallest device that holds it | Decode |
|---|---|---|---|---|---|---|
| 1 | Qwen3-0.6B Qwen · 752M params | 29.7M 1.7K likes | Q4_K_M | 2.22 GB | NVIDIA GeForce GTX 1650 4 GB | ~65 tok/s |
| 2 | Qwen3-VL-8B-Instruct Qwen · 8.8B params | 14.6M 1.2K likes | Q4_K_M | 7.04 GB | NVIDIA GeForce RTX 3070 Ti 8 GB | ~70 tok/s |
| 3 | gemma-4-26B-A4B-it Google · 26B params | 13M 1.6K likes | Q4_K_M | 16.8 GB | AMD Radeon RX 7900 XT 20 GB | ~234 tok/s |
| 4 | Qwen3-8B Qwen · 8.2B params | 10.7M 2.1K likes | Q4_K_M | 7.04 GB | NVIDIA GeForce RTX 3070 Ti 8 GB | ~70 tok/s |
| 5 | gemma-4-31B-it Google · 31B params | 9.9M 4K likes | Q4_K_M | 22.2 GB | NVIDIA GeForce RTX 4090 24 GB | ~33 tok/s |
| 6 | Qwen3.5-9B Qwen · 9.7B params | 9M 2.1K likes | Q4_K_M | 7.05 GB | NVIDIA GeForce RTX 3070 Ti 8 GB | ~85 tok/s |
| 7 | Qwen2.5-0.5B-Instruct Qwen · 494M params | 8.7M 653 likes | Q4_K_M | 1.31 GB | NVIDIA GeForce GTX 1650 4 GB | ~203 tok/s |
| 8 | Qwen2.5-7B-Instruct Qwen · 7.6B params | 8.4M 2.4K likes | Q4_K_M | 5.95 GB | NVIDIA GeForce RTX 2060 6 GB | ~47 tok/s |
| 9 | Qwen3.5-4B Qwen · 4.7B params | 7.8M 1K likes | Q4_K_M | 3.98 GB | NVIDIA GeForce GTX 1650 4 GB | ~32 tok/s |
| 10 | Qwen3-4B Qwen · 4.0B params | 7.8M 720 likes | Q4_K_M | 4.51 GB | NVIDIA GeForce RTX 2060 6 GB | ~62 tok/s |
| 11 | Qwen2.5-1.5B-Instruct Qwen · 1.5B params | 7.4M 863 likes | Q4_K_M | 2.15 GB | NVIDIA GeForce GTX 1650 4 GB | ~72 tok/s |
| 12 | Qwen3.8-27B Qwen · 28B params | 6.9M 16.9K likes | Q4_K_M | 18.6 GB | AMD Radeon RX 7900 XT 20 GB | ~42 tok/s |
| 13 | gpt-oss-20b OpenAI · 21B params | 6.6M 5.1K likes | Q4_K_M | 13.8 GB | AMD Radeon™ RX 9070 XT 16 GB | ~199 tok/s |
| 14 | Qwen2.5-VL-7B-Instruct Qwen · 8.3B params | 5.8M 1.7K likes | Q4_K_M | 6.36 GB | NVIDIA GeForce RTX 3070 Ti 8 GB | ~85 tok/s |
| 15 | GLM-5.3-Flash zai-org · 321B params | 5.4M 2.7K likes | Q4_K_M | 198.3 GB | AMD Instinct MI325X Accelerator 256 GB | ~495 tok/s |
| 16 | Qwen3.5-2B Qwen · 2.3B params | 4.9M 423 likes | Q4_K_M | 2.32 GB | NVIDIA GeForce GTX 1650 4 GB | ~71 tok/s |
| 17 | dolphin-2.9.1-yi-1.5-34b dphn · 34B params | 4.7M 69 likes | Q4_K_M | 23.5 GB | NVIDIA GeForce RTX 4090 24 GB | ~31 tok/s |
| 18 | DeepSeek-V4-Flash-0731 DeepSeek · 304B params | 4.5M 4K likes | Q4_K_M | 187.5 GB | AMD Instinct MI300X Accelerator 192 GB | ~367 tok/s |
| 19 | gpt-oss-120b OpenAI · 117B params | 4.5M 5.3K likes | Q4_K_M | 71.9 GB | NVIDIA RTX PRO 5000 Blackwell 72GB 72 GB | ~257 tok/s |
| 20 | gemma-4-E4B-it Google · 8.0B params | 4.4M 1.6K likes | Q4_K_M | 5.86 GB | NVIDIA GeForce RTX 2060 6 GB | ~76 tok/s |
| 21 | Qwen2.5-3B-Instruct Qwen · 3.1B params | 4.2M 587 likes | Q4_K_M | 3.21 GB | NVIDIA GeForce GTX 1650 4 GB | ~39 tok/s |
| 22 | Qwen3-32B Qwen · 33B params | 4M 754 likes | Q4_K_M | 22.7 GB | NVIDIA GeForce RTX 4090 24 GB | ~32 tok/s |
| 23 | Qwen3-4B-Instruct-2507 Qwen · 4.0B params | 3.7M 989 likes | Q4_K_M | 4.72 GB | NVIDIA GeForce RTX 2060 6 GB | ~62 tok/s |
| 24 | DeepSeek-V3.2 DeepSeek · 685B params | 3.6M 1.5K likes | Q4_K_M | 422.2 GB | AMD Instinct MI455X 432 GB | ~842 tok/s |
| 25 | Qwen3-VL-4B-Instruct Qwen · 4.4B params | 3.5M 485 likes | Q4_K_M | 4.72 GB | NVIDIA GeForce RTX 2060 6 GB | ~62 tok/s |
| 26 | pythia-160m EleutherAI · 213M params | 3.5M 46 likes | Q4_K_M | 1.01 GB | NVIDIA GeForce GTX 1650 4 GB | ~496 tok/s |
| 27 | NVIDIA-Nemotron-3-Nano-4B-BF16 NVIDIA · 4.0B params | 3.4M 126 likes | Q4_K_M | 3.46 GB | NVIDIA GeForce GTX 1650 4 GB | ~33 tok/s |
| 28 | Qwen3.6-35B-A3B Qwen · 36B params | 3.3M 2.9K likes | Q4_K_M | 23.1 GB | NVIDIA GeForce RTX 4090 24 GB | ~370 tok/s |
| 29 | Qwen3-1.7B Qwen · 2.0B params | 3.1M 562 likes | Q4_K_M | 3.02 GB | NVIDIA GeForce GTX 1650 4 GB | ~43 tok/s |
| 30 | gemma-4-E2B-it Google · 5.1B params | 3M 999 likes | Q4_K_M | 4.02 GB | NVIDIA GeForce RTX 2060 6 GB | ~200 tok/s |
| 31 | OTel-2.0-LLM-31B-IT farbodtavakkoli · 31B params | 3M 22 likes | Q4_K_M | 22.2 GB | NVIDIA GeForce RTX 4090 24 GB | ~33 tok/s |
| 32 | Qwen3-VL-2B-Instruct Qwen · 2.1B params | 2.8M 478 likes | Q4_K_M | 3.05 GB | NVIDIA GeForce GTX 1650 4 GB | ~43 tok/s |
| 33 | Qwen3-14B Qwen · 15B params | 2.7M 487 likes | Q4_K_M | 11.1 GB | NVIDIA GeForce RTX 5070 12 GB | ~46 tok/s |
| 34 | Qwen3.5-0.8B Qwen · 873M params | 2.6M 742 likes | Q4_K_M | 1.46 GB | NVIDIA GeForce GTX 1650 4 GB | ~155 tok/s |
| 35 | JiRackUltra_1b CMSManhattan · 1.8B params | 2.5M 0 likes | Q4_K_M | 2.15 GB | NVIDIA GeForce GTX 1650 4 GB | ~72 tok/s |
| 36 | Qwen3.6-27B Qwen · 28B params | 2.5M 2.3K likes | Q4_K_M | 18.6 GB | AMD Radeon RX 7900 XT 20 GB | ~42 tok/s |
| 37 | Qwen2.5-VL-3B-Instruct Qwen · 3.8B params | 2.4M 709 likes | Q4_K_M | 3.41 GB | NVIDIA GeForce GTX 1650 4 GB | ~39 tok/s |
| 38 | Qwen2.5-Coder-7B-Instruct Qwen · 7.6B params | 2.3M 813 likes | Q4_K_M | 5.95 GB | NVIDIA GeForce RTX 2060 6 GB | ~47 tok/s |
| 39 | Mistral-7B-Instruct-v0.3 Mistral AI · 7.2B params | 2.2M 3.7K likes | Q4_K_M | 6.25 GB | NVIDIA GeForce RTX 3070 Ti 8 GB | ~77 tok/s |
| 40 | Qwen2.5-32B-Instruct Qwen · 33B params | 1.9M 362 likes | Q4_K_M | 22.8 GB | NVIDIA GeForce RTX 4090 24 GB | ~32 tok/s |
| 41 | gemma-4-12B-it Google · 12B params | 1.9M 1.6K likes | Q4_K_M | 9.54 GB | NVIDIA GeForce RTX 3080 10 GB | ~63 tok/s |
| 42 | Qwen3.5-27B Qwen · 28B params | 1.9M 1.1K likes | Q4_K_M | 18.6 GB | AMD Radeon RX 7900 XT 20 GB | ~42 tok/s |
| 43 | SmolLM2-135M-Instruct HuggingFaceTB · 135M params | 1.8M 430 likes | Q4_K_M | 1.09 GB | NVIDIA GeForce GTX 1650 4 GB | ~316 tok/s |
| 44 | SmolLM2-135M HuggingFaceTB · 135M params | 1.8M 242 likes | Q4_K_M | 1.09 GB | NVIDIA GeForce GTX 1650 4 GB | ~316 tok/s |
| 45 | GLM-4.7-Flash zai-org · 31B params | 1.8M 1.9K likes | Q4_K_M | 20.4 GB | NVIDIA GeForce RTX 4090 24 GB | ~35 tok/s |
| 46 | Mistral-7B-Instruct-v0.2 Mistral AI · 7.2B params | 1.7M 3.2K likes | Q4_K_M | 6.24 GB | NVIDIA GeForce RTX 3070 Ti 8 GB | ~77 tok/s |
| 47 | Qwen2.5-Coder-14B-Instruct Qwen · 15B params | 1.7M 192 likes | Q4_K_M | 11.4 GB | NVIDIA GeForce RTX 5070 12 GB | ~45 tok/s |
| 48 | Qwen2.5-14B-Instruct Qwen · 15B params | 1.6M 372 likes | Q4_K_M | 11.4 GB | NVIDIA GeForce RTX 5070 12 GB | ~45 tok/s |
| 49 | Qwen3.5-35B-A3B Qwen · 36B params | 1.6M 1.5K likes | Q4_K_M | 23.1 GB | NVIDIA GeForce RTX 4090 24 GB | ~370 tok/s |
| 50 | Qwen3-30B-A3B Qwen · 31B params | 1.5M 953 likes | Q4_K_M | 20.2 GB | NVIDIA GeForce RTX 4090 24 GB | ~251 tok/s |
How the requirement column is computed
Each row is a real run of the same engine the calculator uses: weights at the quantization shown, the KV cache at 8,192 tokens in F16, and the runtime reserve — with offload switched off, so “holds it” means fully resident rather than technically loadable.
Devices are ordered by published capacity, not by price. This site has no price data, and ranking by an invented one would be exactly the kind of unsourced figure it refuses everywhere else.
Looking at it from the other side — what does my card run? — every hardware page answers that for one device.