Can I run this LLM? Requirements for every model, measured from real files.
89 models × every quantization × every GPU. Sizes are pulled daily from Hugging Face GGUF repos — not computed from formulas, so what you see is what you download.
What LLM can I run?interactive picker — type your VRAM
Best LLM for 24GB VRAMranked, measured
Best LLM for RTX 409024GB, ranked
Best LLM for RTX 306012GB, ranked
Best LLM for 8GB VRAMranked, measured
Guides: VRAM, quants, runtimeshow it all works
Which GPU should I buy?cheapest hardware per model
Head-to-head comparisons
Kimi-K3 vs deepseek-v4measured hardware comparisondeepseek-v4 vs gemma-4-12b-itmeasured hardware comparisonKimi-K3 vs Qwen3.8-27Bmeasured hardware comparisonGLM-5.3 vs Qwen3.8-27Bmeasured hardware comparison
Qwen3.8-Flash-Next-GSQ-RCO? params · smallest build 81.2 GB · 6513k downloadsQwen3.8-27B27B params · smallest build 9.3 GB · 5751k downloadsQwen3.8-27B-Uncensored27B params · smallest build 3.5 GB · 2405k downloadsQwen-Image-2.1-Uncensored-Quantized7B params · smallest build 6.5 GB · 2398k downloadsHuihui-Qwen3.8-27B-abliterated27B params · smallest build 9.4 GB · 2388k downloadsTernary-Bonsai-2-27B27B params · smallest build 2.3 GB · 2336k downloadsQwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP27B params · smallest build 13.9 GB · 2277k downloadsQwen3.8-Flash-Next-GSQ-RCO-abliterated? params · smallest build 81.7 GB · 2117k downloadsQwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NEO-CODER-MAX-MTP27B params · smallest build 30.1 GB · 1679k downloadsgemma-4-12b-it12B params · smallest build 6.6 GB · 1499k downloadsQwen3.5-4B4B params · smallest build 3.3 GB · 1232k downloadsQwen3.8-27B-GSQ-RCO27B params · smallest build 22.1 GB · 1226k downloadsdeepseek-v4403B params · smallest build 118.6 GB · 1126k downloadsgemma-4-26B-A4B-it25B params · smallest build 12.9 GB · 963k downloadsQwen3-VL-8B-Instruct8B params · smallest build 7.5 GB · 945k downloadsGemma-4-E4B-Uncensored-HauhauCS-Aggressive4B params · smallest build 6.8 GB · 920k downloadsGLM-5.3-Flash321B params · smallest build 113.2 GB · 857k downloadsQwen3-4B4B params · smallest build 4.5 GB · 584k downloadsQwen3-8B8B params · smallest build 7.5 GB · 569k downloadsQwen3.8-4B-Distill4B params · smallest build 4.8 GB · 541k downloadsQwen-Image-2.1-viggle-turbo7B params · smallest build 11.9 GB · 521k downloadsQwen3.8-27B-Uncensored-Cyber27B params · smallest build 24.3 GB · 507k downloadsTernary-Bonsai-27B27B params · smallest build 2.3 GB · 501k downloadsorcarouter_GLM-5.3-Flash-Uncensored? params · smallest build 118.0 GB · 491k downloadsMeta-Llama-3.1-8B-Instruct8B params · smallest build 5.0 GB · 488k downloadsQwen-Image-2.1-Turbo-Uncensored8B params · smallest build 5.5 GB · 468k downloadsQwen3-VL-4B-Instruct4B params · smallest build 4.5 GB · 462k downloadsGLM-5.3754B params · smallest build 261.6 GB · 372k downloadsgemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau212B params · smallest build 8.8 GB · 344k downloadsQwen3.8-27B-Turbo-Brilliance-50X-Reasoning-Instruct-NEO-MAX-MTP27B params · smallest build 16.0 GB · 343k downloadsHuihui-DeepSeek-V4-Flash-0731-abliterated? params · smallest build 105.6 GB · 323k downloadsQwen3.6-14B-A3B-FableVibes0B params · smallest build 2.6 GB · 317k downloadsKimi-K3? params · smallest build 561.1 GB · 287k downloadsgranite-4.2-3b3B params · smallest build 3.2 GB · 254k downloadsgranite-4.2-8b8B params · smallest build 5.6 GB · 251k downloadsQwen3-14B14B params · smallest build 12.3 GB · 246k downloadsgranite-4.2-30b30B params · smallest build 14.5 GB · 230k downloadsQwen2.5-VL-7B-Instruct8B params · smallest build 4.0 GB · 223k downloadsMiniCPM-V-4.7-35B-A3B35B params · smallest build 18.9 GB · 218k downloadsParable-Qwen3-8B-Claude-Fable-58B params · smallest build 7.5 GB · 215k downloadsQwen2.5-7B-Instruct7B params · smallest build 4.8 GB · 204k downloadsKimi-K2.7-Code? params · smallest build 328.8 GB · 192k downloadsGLM-5.2754B params · smallest build 261.6 GB · 179k downloadsLFM2.5-24B-A2B24B params · smallest build 17.8 GB · 165k downloadsParable-Granite-4.1-3B-Claude-Fable-53B params · smallest build 4.0 GB · 161k downloadsParable-Granite-4.1-8B-Claude-Fable-58B params · smallest build 7.6 GB · 159k downloadsLlama-3.2-3B-Instruct3B params · smallest build 3.4 GB · 159k downloadshumanizer12B params · smallest build 8.1 GB · 129k downloadsembeddinggemma-20B params · smallest build 1.7 GB · 121k downloadsDeepSeek-V4-Flash-0731? params · smallest build 14.6 GB · 118k downloadsPhi-4-mini-instruct4B params · smallest build 3.5 GB · 104k downloadsgemma-3-4b-it4B params · smallest build 3.3 GB · 100k downloadsmicrosoft_Phi-4-mini-instruct? params · smallest build 3.3 GB · 95k downloadsgemma-3-12b-it-qat12B params · smallest build 7.3 GB · 93k downloadsMistral-Small-Instruct-2409? params · smallest build 7.3 GB · 84k downloadsXing4.0-29B-A4B29B params · smallest build 11.3 GB · 64k downloadsgemma-3-12b-it12B params · smallest build 7.3 GB · 62k downloadsMistral-Small-3.2-24B-Instruct-250624B params · smallest build 8.2 GB · 57k downloadsMinistral-3-3B-Instruct-25124B params · smallest build 4.1 GB · 50k downloadsLFM2.5-24B-A2B-Uncensored24B params · smallest build 16.0 GB · 48k downloadsQwen-Image-2.1-Uncensored-Abenzerps7B params · smallest build 6.5 GB · 48k downloadsMinistral-3-8B-Instruct-25129B params · smallest build 7.7 GB · 43k downloadsmistralai_Mistral-Small-3.2-24B-Instruct-250624B params · smallest build 9.4 GB · 42k downloadsLlama-3.2-3B-Instruct-uncensored3B params · smallest build 3.3 GB · 42k downloadsQwen-Image-2.1-Uncensored? params · smallest build 6.4 GB · 39k downloadsgemma-3n-E4B-it4B params · smallest build 4.9 GB · 35k downloadsMiniCPM-V-4.7-35B-A3B-CRACK35B params · smallest build 21.8 GB · 34k downloadsflint40B params · smallest build 15.1 GB · 33k downloadsSakura-Qwen-Image-2.1-Turbo-Uncensored? params · smallest build 6.4 GB · 31k downloadsMinistral-3-14B-Instruct-251214B params · smallest build 11.4 GB · 30k downloadsDeepSeek-V4.1-Flash-1017? params · smallest build 281.1 GB · 29k downloadsKimi-K2-Instruct1027B params · smallest build 263.7 GB · 15k downloadsQwen3.5-9B-Kimi-k3-Distilled-i19B params · smallest build 6.2 GB · 13k downloadsMiniCPM-V-4.7-35B-A3B-i135B params · smallest build 10.5 GB · 12k downloadshumanizer-i1? params · smallest build 5.1 GB · 9k downloadsMiniCPM-V-4.7-35B-A3B-Uncensored35B params · smallest build 25.7 GB · 8k downloadsLFM2.5-24B-A2B-Turbo-Brilliance-50X-NEO-MAX-Imatrix24B params · smallest build 14.1 GB · 7k downloadsKimi-K2.51027B params · smallest build 288.9 GB · 7k downloadsKimi-K2-Thinking1027B params · smallest build 297.7 GB · 6k downloadsViggle_Qwen-Image-2.1-Turbo_GGUFs? params · smallest build 6.0 GB · 5k downloadsKimi-K2.61027B params · smallest build 410.1 GB · 5k downloadsHumanizer-12B12B params · smallest build 4.4 GB · 4k downloadsXing4.0-29B-A4B-CRACK29B params · smallest build 18.1 GB · 3k downloadsSakura-EmbeddingGemma-2-HighQuality? params · smallest build 1.7 GB · 1k downloadsSpark-X2.6-4B4B params · smallest build 3.4 GB · 1k downloadsSpark-X2.6-4B-Tema_Q-Abliterated-i14B params · smallest build 3.0 GB · 1k downloadsSpark-X2.6-4B-i14B params · smallest build 3.0 GB · 0k downloadsSpark-X2.6-4B-abliterated-FIT4B params · smallest build 4.3 GB · 0k downloadsDeepSeek-V4.1-Flash-1017-GSQ-RCO? params · smallest build 134.2 GB · 0k downloads
Frequently asked questions
How do I know if I can run an LLM on my PC?
Compare your VRAM (GPU) or system RAM (CPU inference) with the model's GGUF file size plus KV-cache overhead. Rough guide: an 8B model at Q4_K_M needs ~6 GB VRAM at 8K context, a 27-30B model needs ~18-20 GB, and 70B+ needs multi-GPU or CPU offload. Every ModelFit page lists the measured file size for each quantization so you can match it to your exact hardware.
Where do ModelFit's numbers come from?
Directly from the Hugging Face API, refreshed daily: the actual published GGUF file sizes from community repos (unsloth, Bartowski and others) plus KV-cache and runtime overhead for 8K context. Nothing is estimated from parameter-count formulas.
Which quantization should I pick?
Q4_K_M is the standard balance of quality and size. Drop to IQ3/Q3 only when VRAM is tight; choose Q5/Q6 when you have headroom and want maximum quality. Each model page shows what every quantization needs on your hardware.