What LLMs fit in 32GB VRAM? (Oct 2026)
27 models on our list fit in 32 GB of graphics memory. The largest by parameter count is Kolibri-1 at Q2_K, 30.93 GB (measured weights; total is an estimate). We keep 0.6 GB free on one graphics card, so 31.4 GB is usable. Each total adds conversation memory for 8,192 tokens (or the model's own maximum, when that is shorter) and 0.8 GB for the software. The total is an estimate.
Updated
Real-world 32 GB workstation card report [self-reported], 24 Oct 2026: on one Intel Arc Pro B70 32GB, Josh Gold (github.com/killercatfish, llama.cpp #30892, quoted with permission) reports about 30 tokens per second serving Qwen3.8-27B Int4 (AutoRound) on vLLM, and about 15 tokens per second generation with about 375–500 tokens per second of prompt reading on llama.cpp Vulkan at Q4_K_M. The same 32 GB card also holds a Q6 file for long editing jobs (about 23.3 GB at 8,192 tokens, our estimate). Can I run Qwen3.8-27B. Intel Arc Pro B70 32GB. Josh Gold's interview.
See also Gabe Ortiz's RTX 5090 32 GB workstation interview (~220 tok/s with DFlash2 speculative decoding on Qwen3.8-27B 4-bit 17.00 GB, self-reported). RTX 5090 32GB · Gabe Ortiz's interview.
Own a 32 GB GPU? Get a 1-line alert when a new top model fits 31.4 GB usable VRAM
27 models on our catalog fit 32 GB today. Get our weekly 1-line alert when a new open-weight LLM lands inside 31.4 GB at Q4_K_M or when 32 GB+ GPU prices drop >10% — zero spam.
Models that fit
Ranked by parameter count, largest first. The best quant is the highest-quality option that fits in 31.4 GB. A measured download is used when we have one. Otherwise the figure is Q4, Q5, Q8 or FP16 from our estimate. The gigabytes are the weights plus conversation memory for 8,192 tokens, or the model's shorter maximum, plus 0.8 GB for the software. The total is an estimate.
| Rank | AI model | Parameters | Best quant | GB | Label |
|---|---|---|---|---|---|
| 1 | Can I run Kolibri-1? | 78.1B (estimate), 3.46B active | Q2_K | 30.93 GB | measured weights; total is an estimate |
| 2 | Can I run Recursive Flint? | 39.79B, 2.5B active | INT4-Q4_1 | 27.06 GB | measured weights; total is an estimate |
| 3 | Can I run Qwen3.6-35B-A3B? | 35.95B (estimate), 3B active | Q5 (~0.71 bytes/param) | 26.49 GB | estimate |
| 4 | Can I run Nanbeige4.3-33B? | 33.12B (estimate) | Q5_K_M | 27.61 GB | measured weights; total is an estimate |
| 5 | Can I run Qwen2.5 32B? | 32.8B (estimate) | Q5 (~0.71 bytes/param) | 26.24 GB | estimate |
| 6 | Can I run Qwen3 32B? | 32.76B (estimate) | Q6_K | 29.83 GB | measured weights; total is an estimate |
| 7 | Can I run Gemma 4 31B? | 31.27B (estimate) | Q6_K | 27.51 GB | measured weights; total is an estimate |
| 8 | Can I run Qwen3 Coder 30B A3B? | 30.53B (estimate), 3.3B active | Q6_K | 26.70 GB | measured weights; total is an estimate |
| 9 | Can I run NVIDIA Nemotron 3.5 Lightning? | 30B (estimate), 3B active | UD-Q5_K_XL | 31.26 GB | measured weights; total is an estimate |
| 10 | Can I run Muse Glimmer 30B? | 29.6B | KQuant-Dynamic-Q4_K_XL | 20.64 GB | measured weights; total is an estimate |
| 11 | Can I run Qwen3.8-27B? | 27.78B (estimate) | Q8_0 | 30.39 GB | measured weights; total is an estimate |
| 12 | Can I run Gemma 3 27B? | 27.43B (estimate) | Q8_0 | 30.62 GB | measured weights; total is an estimate |
| 13 | Can I run Gemma 2 27B? | 27.2B (estimate) | Q5 (~0.71 bytes/param) | 22.43 GB | estimate |
| 14 | Can I run OrcaCyber Nano 1.0? | 27B | Q8 | 30.23 GB | estimate |
| 15 | Can I run Gemma 4 26B A4B? | 25.2B (estimate), 3.8B active | Q8_0 | 28.04 GB | measured weights; total is an estimate |
| 16 | Can I run Mistral Small 3.2 24B? | 24.01B (estimate) | Q8_0 | 27.19 GB | measured weights; total is an estimate |
| 17 | Can I run gpt-oss-20b? | 21B (estimate), 3.6B active | F16 | 14.79 GB | measured weights; total is an estimate |
| 18 | Can I run DeepSeek R1 Distill Qwen 14B? | 14.77B | Q8_0 | 18.11 GB | measured weights; total is an estimate |
| 19 | Can I run Phi-4 14B? | 14.7B (estimate) | Q8 | 18.21 GB | estimate |
| 20 | Can I run Qwen2.5 14B? | 14.7B (estimate) | Q8 | 18.14 GB | estimate |
| 21 | Can I run Gemma 2 9B? | 9.24B (estimate) | FP16 | 21.39 GB | estimate |
| 22 | Can I run Llama 3.1 8B? | 8.03B (estimate) | FP16 | 17.93 GB | estimate |
| 23 | Can I run NVIDIA Nemotron 3.5 Bolt E4B? | 8B (estimate) | BF16 | 16.12 GB | measured weights; total is an estimate |
| 24 | Can I run Mistral 7B v0.3? | 7.25B (estimate) | FP16 | 16.37 GB | estimate |
| 25 | Can I run Spark-X2.6-4B? | 4.11B (estimate) | BF16 | 9.39 GB | measured weights; total is an estimate |
| 26 | Can I run Llama 3.2 3B? | 3.21B (estimate) | FP16 | 8.16 GB | estimate |
| 27 | Can I run LFM3-1.7B? | 1.73B (estimate) | BF16 | 4.38 GB | measured weights; total is an estimate |
Cards and Macs with 32 GB
The euro price is the figure we store, with the date it was written down. A Mac's gigabytes are shared memory, not graphics memory, so a Mac does not hold the same models as a 32 GB card.
| Machine | Memory | EU price | Date | Label |
|---|---|---|---|---|
| GeForce RTX 5090 32GB | 32 GB graphics memory | €7,367 | New price, the lowest in stock in the EU | |
| Intel Arc Pro B70 32GB | 32 GB graphics memory | No euro price recorded | no euro price recorded | |
| Mac mini M6 32GB | 32 GB shared memory | €1,530 | price unverified |
Memory totals are our estimate, including when the weight file was measured. Prices are the euro figures we store, with the date they were written down. How we calculate
Common questions
- What is the largest model that fits in 32 GB of VRAM?
- Kolibri-1 is the largest by parameter count, at Q2_K, 30.93 GB (measured weights; total is an estimate). 27 models on our list fit.
- What does the best quant mean for 32 GB?
- The best quant is the highest-quality option that still fits. A measured download is used when we have one at that quality. Otherwise the figure is Q4, Q5, Q8 or FP16 from our estimate. A higher bit count wins. A file that runs only on a Mac is not counted. The gigabytes add conversation memory and the software overhead, so the total is an estimate even when the weight file was measured.
- Which cards and Macs have 32 GB, and what do they cost in the EU?
- GeForce RTX 5090 32GB (32 GB graphics memory): €7,367, New price, the lowest in stock in the EU, 21 Oct 2026. Intel Arc Pro B70 32GB (32 GB graphics memory): No euro price recorded, no euro price recorded, 24 Oct 2026. Mac mini M6 32GB (32 GB shared memory): €1,530, price unverified, 16 Oct 2026.
- How much of 32 GB can a model use?
- Our estimate keeps 0.6 GB free on one graphics card, so 31.4 GB is usable. A Mac with 32 GB of shared memory does not give the graphics all of it. The model table is for one graphics card.