Most popular local LLMs
Ranked by real download counts, refreshed weekly. Data fetched 2026-10-20.
Ranking
Top on this site right now: Qwen3.8 27B, Gemma 4 26B-A4B, Qwen3.5 4B, Gemma 4 31B, Qwen3.5 9B.
| # | Model | Params | Score (30-day downloads) | Original repo | GGUF repos | Likes | Ollama (all-time, family) |
|---|---|---|---|---|---|---|---|
| 1 | Qwen3.8 27B | 27.8B | 17.7M | 7.6M | 6.4M + 294K + 2.2M + 1.1M | 18K | 4.5M qwen3.8 |
| 2 | Gemma 4 26B-A4B | 25.8B | 11.4M | 10.1M | 920K + 66K + 342K + 42K | 2K | 27.4M gemma4 |
| 3 | Qwen3.5 4B | 4.7B | 9.4M | 7.8M | 1.3M + 127K + 153K | 1K | 22.9M qwen3.5 |
| 4 | Gemma 4 31B | 31.3B | 9.3M | 8.9M | 280K + 35K + 35K + 37K | 4K | 27.4M gemma4 |
| 5 | Qwen3.5 9B | 9.7B | 9.2M | 7.3M | 1.3M + 163K + 482K | 2K | 22.9M qwen3.5 |
| 6 | zai-org/GLM-5.3-Flash (not on the site) | 321.3B | 7.5M | 6.5M | 935K + 7K | 3K | 190.4K glm-5.3-flash |
| 7 | gpt-oss 20B | 20.9B | 6.8M | 5.9M | 536K + 13K + 217K + 94K | 5K | 14M gpt-oss |
| 8 | Qwen3-VL 8B | 8.8B | 6.6M | 5.4M | 208K + 24K + 56K + 916K | 1K | 6.8M qwen3-vl |
| 9 | Ornith 1.5 9B | 9.7B | 5.5M | 448K | 22K + 5.0M | 386 | - |
| 10 | Qwen3.6 35B-A3B | 36B | 5.3M | 3.6M | 1.3M + 180K + 121K + 130K | 3K | 7.2M qwen3.6 |
| 11 | Gemma 4 E4B | 8B | 5.3M | 4.1M | 611K + 38K + 123K + 432K | 2K | 27.4M gemma4 |
| 12 | Ornith 1.5 35B-A3B | 36B | 4.1M | 257K | 48K + 3.8M | 744 | - |
| 13 | Qwen3.8 Flash Next | 180B | 3.8M | 2.5M | 1.1M + 69K + 44K + 12K | 6K | 4.5M qwen3.8 |
| 14 | Gemma 4 12B | 12B | 3.5M | 1.7M | 1.6M + 39K + 63K + 111K | 2K | 27.4M gemma4 |
| 15 | gpt-oss 120B | 116.8B | 3.4M | 3.1M | 79K + 5K + 199K + 19K | 5K | 14M gpt-oss |
| 16 | Qwen3.6 27B | 27.8B | 2.9M | 2.3M | 430K + 42K + 79K + 63K | 2K | 7.2M qwen3.6 |
| 17 | Qwen/Qwen3.5-35B-A3B (not on the site) | 36B | 2.5M | 1.4M | 959K + 41K + 5K + 51K | 2K | 22.9M qwen3.5 |
| 18 | Nemotron 3 Nano 4B | 4B | 1.8M | 1.7M | 7K + 42K + 14K | 138 | 929.9K nemotron-3-nano |
| 19 | Qwen3-Coder 30B-A3B | 30.5B | 1.7M | 552K | 1.1M + 77K | 1K | 10.2M qwen3-coder |
| 20 | thinkingmachines/Inkling-Small (not on the site) | 266B | 1.5M | 786K | 750K | 423 | - |
| 21 | deepseek-ai/DeepSeek-V4.1-Flash (not on the site) | 763.2B | 1.4M | 1.4M | - | 5K | - |
| 22 | GLM-4.7-Flash | 31.2B | 1.4M | 1.1M | 231K + 29K + 30K + 29K | 2K | 2.1M glm-4.7-flash |
| 23 | Nemotron 3 Nano 30B-A3B | 31.6B | 1.2M | 1.2M | 5K + 4K | 831 | 929.9K nemotron-3-nano |
| 24 | Nemotron 3.5 Lightning 30B-A3B | 31.6B | 961K | 695K | 46K + 31K + 187K + 2K | 243 | - |
| 25 | GLM-4.5-Air | 110.5B | 793K | 764K | 24K + 4K + 718 + 727 | 646 | - |
| 26 | Laguna XS 2.1 | 33.4B | 650K | 441K | 7K + 2K + 1K + 198K | 257 | - |
| 27 | DeepSeek R1 Distill Qwen 7B | 7.6B | 554K | 386K | 48K + 107K + 13K | 936 | 93.9M deepseek-r1 |
| 28 | DeepSeek R1 Distill Qwen 32B | 32.8B | 451K | 363K | 24K + 55K + 9K | 2K | 93.9M deepseek-r1 |
| 29 | deepseek-ai/DeepSeek-R1-Distill-Qwen-14B (not on the site) | 14.8B | 435K | 328K | 33K + 49K + 25K | 721 | 93.9M deepseek-r1 |
| 30 | microsoft/phi-4 (not on the site) | 14.7B | 411K | 373K | 10K + 10K + 9K + 10K | 2K | 7.8M phi4 |
| 31 | mistralai/Devstral-Small-2-24B-Instruct-2512 (not on the site) | 24B | 381K | 253K | 62K + 23K + 2K + 41K | 702 | 2M devstral-small-2 |
| 32 | Llama 3.3 70B | 70.6B | 324K | 234K | 34K + 41K + 15K | 3K | 4.3M llama3.3 |
| 33 | Mistral Small 3.2 24B | 24B | 297K | 207K | 57K + 27K + 6K | 638 | 2.5M mistral-small3.2 |
| 34 | Kwaipilot/KAT-Coder-V2.5-Dev (not on the site) | 34.7B | 269K | 12K | 257K | 669 | - |
| 35 | LFM2.5 24B-A2B | 24B | 205K | 15K | 50K + 5K + 1K + 133K | 524 | - |
| 36 | deepseek-ai/DeepSeek-R1-Distill-Llama-8B (not on the site) | 8B | 186K | 128K | 43K + 11K + 5K | 911 | 93.9M deepseek-r1 |
| 37 | Nemotron 3.5 Super VL 120B-A12B | 124.5B | 91K | 19K | 14K + 32K + 27K | 82 | - |
| 38 | deepseek-ai/DeepSeek-R1-Distill-Llama-70B (not on the site) | 70.6B | 90K | 43K | 33K + 11K + 4K | 819 | 93.9M deepseek-r1 |
Method and sources
Score = Hugging Face downloads over the last 30 days for the original repo plus its plain GGUF conversions from unsloth, bartowski, ggml-org, lmstudio-community and the model maker's own GGUF repo. Likes and Ollama library downloads (all-time, per model family) are shown for context and are not part of the score. The Hugging Face number is the API's downloads field, which model pages show as "Downloads last month". How the Hub counts them (every request to a counted file, including each GGUF file, so whole-repo clones count more than once) is described in the HF download-stats docs. Every number links to the repo it came from. Ollama counts come from each ollama.com library page; they are all-time and shared by every size in a family (for example all DeepSeek R1 distills share one count), so they are shown for context only. Download counts include automated pulls and CI, so read the ranking as relative demand, not users. Candidates are the site's models plus other popular open-weight models we considered; models above about 250B parameters or without GGUF files are not added to the site. Fetched 2026-10-20 by tools/demand_rank.py.