Best local LLMs for 32GB VRAM
What you can run with 32 GB of GPU memory (31 GB usable after ~1 GB overhead), at 16k context with an f16 KV cache. Speeds use data measured on GeForce RTX 5090 where available.
Models that fit 32 GB, largest first
| Model | Params | Quant | Fit | Memory | Tok/s | Speed source |
|---|---|---|---|---|---|---|
| Qwen3.6 35B-A3B | 35.95B MoE | UD-Q6_K | fits fully | 29.7 GB | 260 | measured |
| Gemma 4 31B | 31.27B | Q6_K | fits fully | 27.4 GB | 55 | measured |
| Qwen3-Coder 30B-A3B | 30.53B MoE | Q6_K | fits fully | 26.7 GB | 316 | scaled from measured |
| Qwen3.6 27B | 27.78B | Q8_0 | fits fully | 29.8 GB | 49 | scaled from measured |
| Gemma 4 26B-A4B | 25.81B MoE | Q8_0 | fits fully | 27.9 GB | 213 | scaled from measured |
| Mistral Small 3.2 24B | 24.01B | Q8_0 | fits fully | 27.7 GB | 56 | scaled from measured |
| gpt-oss 20B | 20.91B MoE | Q8_0 | fits fully | 12.5 GB | 298 | scaled from measured |
| Gemma 4 12B | 11.96B | Q8_0 | fits fully | 13.7 GB | 91 | scaled from measured |
| Qwen3.5 9B | 9.65B | Q8_0 | fits fully | 10.1 GB | 110 | scaled from measured |
| Qwen3-VL 8B | 8.77B | Q8_0 | fits fully | 11.1 GB | 43 | scaled from measured |
| Qwen3.5 4B | 4.66B | Q8_0 | fits fully | 5.1 GB | 72 | scaled from measured |
Copy-paste commands for the top pick
Qwen3.6 35B-A3B UD-Q6_K, 16k context.
llama-server -hf unsloth/Qwen3.6-35B-A3B-GGUF:UD-Q6_K -c 16384 -fa on --jinja -ngl 99
OLLAMA_CONTEXT_LENGTH=16384 ollama run hf.co/unsloth/Qwen3.6-35B-A3B-GGUF:UD-Q6_K
How much context fits in 32 GB?
Largest context where weights plus KV cache stay inside 31 GB, by KV cache type (q8_0 halves and q4_0 quarters the cache). Native limits and YaRN extension apply per model.
| Model | Weights | f16 KV | q8_0 KV | q4_0 KV |
|---|---|---|---|---|
| Qwen3.6 35B-A3B UD-Q6_K | 29.3 GB | 64k | 128k | 256k |
| Gemma 4 31B Q6_K | 25.2 GB | 32k | 64k | 128k |
| Qwen3-Coder 30B-A3B Q6_K | 25.1 GB | 32k | 64k | 128k |
| Qwen3.6 27B Q8_0 | 28.6 GB | 32k | 32k | 64k |
| Gemma 4 26B-A4B Q8_0 | 27.3 GB | 128k | 256k | 256k |
| Mistral Small 3.2 24B Q8_0 | 25.1 GB | 32k | 64k | 64k |
| gpt-oss 20B Q8_0 | 12.1 GB | 128k | 128k | 128k |
| Gemma 4 12B Q8_0 | 13.1 GB | 256k | 256k | 256k |
Try the exact numbers for your setup in the interactive advisor.
What does not fit fully in 32 GB
Every model in the dataset fits.
GPUs with 32 GB
GeForce RTX 5090. See their pages: RTX 5090 32GB.
Hardware pages near 32 GB
RTX 3090 24GB · RTX 4090 24GB · RTX 5090 32GB · AMD RX 7900 XTX 24GB · Mac mini M6 32GB · MacBook Air M5 32GB
FAQ
What is the best local LLM for 32GB VRAM?
By size, Qwen3.6 35B-A3B UD-Q6_K is the largest model in our dataset that fits fully at 16k context (29.7 GB, ~260 tok/s (measured)).
How many models fit in 32GB VRAM?
11 of 13 models fit fully at 16k context with an f16 KV cache; 0 more run only with CPU offload.
How much context can I use with 32GB?
It depends on the model: for Qwen3.6 35B-A3B the KV cache allows up to 64k tokens at f16 and 256k with a q4_0 cache.
Neighbouring tiers: 24 GB · 48 GB · all tiers. Numbers come from the same memory-fit math as the advisor; sources and methods.