Best local AI models for 24GB VRAM
Ranked by what actually fits at 32K context, computed from real file bytes.
A 24GB card gives you about 22.32 GiB to work with after driver overhead. 171 indexed models fit at 32K context — the largest being Qwen3.6-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking at 39.5B parameters in IQ3_M.
From the file· fit from summed bytesFrom the file· KV per layer
Fits in 24GB at 32K context
largest quantization that fits, per model
| Model | Modality | Best quant | Params○ | Total◐ | Headroom◐ |
|---|---|---|---|---|---|
| Qwen3.6-35B-A3BMoE | vision + language | Q4_K_M | 36.0B | 22.18 GiB | 0.14 GiB |
| Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP | vision + language | I1-Q5_K_M | 27.8B | 21.06 GiB | 1.26 GiB |
| Qwen3.5-9B | vision + language | BF16 | 9.7B | 18.98 GiB | 3.34 GiB |
| gemma-4-26B-A4B-itMoE | vision + language | UD-Q5_K_M | 26.5B | 22.03 GiB | 0.29 GiB |
| gemma-4-12B-it | vision + language | Q8_0 | 12.0B | 15.55 GiB | 6.77 GiB |
| Qwen3.5-4B | vision + language | BF16 | 4.7B | 9.88 GiB | 12.44 GiB |
| Qwythos-9B-Claude-Mythos-5-1M | vision + language | Q8_0 | 9.4B | 19.82 GiB | 2.50 GiB |
| gemma-4-31B-it | vision + language | Q3_K_M | 31.3B | 21.88 GiB | 0.44 GiB |
| Muse-Glimmer-30B | vision + language | Q5_K_L | 29.8B | 20.88 GiB | 1.44 GiB |
| gemma-4-26B-A4B-it-qat-q4_0-unquantizedMoE | vision + language | Q4_0 | 26.5B | 15.78 GiB | 6.54 GiB |
| Qwen3.5-0.8B | vision + language | BF16 | 873M | 2.60 GiB | 19.72 GiB |
| gemma-4-E2B-it | vision + language | BF16 | 5.1B | 9.71 GiB | 12.61 GiB |
| gemma-4-E4B-it-qat-q4_0-unquantized | vision + language | Q4_0 | 7.9B | 6.12 GiB | 16.20 GiB |
| Qwen3-VL-30B-A3B-InstructMoE | vision + language | Q4_1 | 31.1B | 21.67 GiB | 0.65 GiB |
| Qwopus3.6-35B-A3B-v1MoE | vision + language | I1-Q4_1 | 36.0B | 21.78 GiB | 0.54 GiB |
| Qwen3.5-9B-The-Defiant-Fable-Uncensored-Heretic-NEO-IMATRIX-MAX-MTP | vision + language | Q8_0 | 9.7B | 21.60 GiB | 0.72 GiB |
| Qwen3.5-35B-A3BMoE | vision + language | Q4_K_M | 36.0B | 22.18 GiB | 0.14 GiB |
| Qwythos-9B-v2 | vision + language | Q8_0 | 9.7B | 19.82 GiB | 2.50 GiB |
| ThinkingCap-Qwen3.6-27B | vision + language | Q5_K_M | 27.4B | 22.19 GiB | 0.13 GiB |
| Qwopus3.6-27B-Coder | vision + language | Q5_K_M | 27.8B | 21.06 GiB | 1.26 GiB |
| Qwen3-VL-4B-Instruct | vision + language | BF16 | 4.4B | 12.81 GiB | 9.51 GiB |
| Qwen3.6-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking | vision + language | IQ3_M | 39.5B | 20.98 GiB | 1.34 GiB |
| Qwen2.5-VL-7B-Instruct | vision + language | BF16 | 8.3B | 16.80 GiB | 5.52 GiB |
| Qwen3-VL-2B-Instruct | vision + language | BF16 | 2.1B | 7.50 GiB | 14.82 GiB |
| Qwen3.5-27B | vision + language | Q5_K_S | 27.8B | 21.39 GiB | 0.93 GiB |
| gemma-3-12b-it | vision + language | Q8_0 | 12.2B | 14.97 GiB | 7.35 GiB |
| Qwen3.5-2B | vision + language | BF16 | 2.3B | 4.80 GiB | 17.52 GiB |
| Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking | vision + language | Q5_K_M | 27.4B | 21.06 GiB | 1.26 GiB |
| Qwen3.6-35B-A3B-Claude-4.7-Opus-Reasoning-DistilledMoE | vision + language | Q4_K | 36.0B | 21.65 GiB | 0.67 GiB |
| Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-PreservedMoE | vision + language | Q4_K_M | 35.1B | 21.71 GiB | 0.61 GiB |
| Qwen3.5-9B | vision + language | Q8_0 | 9.7B | 10.95 GiB | 11.37 GiB |
| Qwen3.6-35B-A3B-uncensored-hereticMoE | vision + language | Q4_K_M | 35.1B | 21.20 GiB | 1.12 GiB |
| gemma-4-31B-it-uncensored-heretic | vision + language | Q3_K_M | 31.3B | 21.29 GiB | 1.03 GiB |
| Qwopus3.6-27B-v2 | vision + language | Q5_K_M | 27.8B | 21.06 GiB | 1.26 GiB |
| Qwen3.6-27B-uncensored-heretic-v2-Native-MTP-Preserved | vision + language | Q5_K_M | 27.4B | 21.25 GiB | 1.07 GiB |
| Mistral-Small-3.2-24B-Instruct-2506 | vision + language | Q5_K_L | 24.0B | 21.92 GiB | 0.40 GiB |
| LFM2.5-VL-1.6B | vision + language | BF16 | 1.6B | 3.37 GiB | 18.95 GiB |
| gemma-4-E4B-it-ultra-uncensored-heretic | vision + language | BF16 | 8.0B | 15.34 GiB | 6.98 GiB |
| Unlimited-OCRMoE | vision + language | BF16 | 3.3B | 8.14 GiB | 14.18 GiB |
| diffusiongemma-26B-A4B-itMoE | vision + language | Q5_K_M | 25.8B | 20.16 GiB | 2.16 GiB |
| Qwopus3.5-9B-v3.5 | vision + language | BF16 | 9.7B | 18.53 GiB | 3.79 GiB |
| Qwen3.6-27B-uncensored-heretic-v2 | vision + language | Q5_K_M | 27.4B | 20.77 GiB | 1.55 GiB |
| Tess-4-27B | vision + language | Q5_K_M | 27.8B | 22.19 GiB | 0.13 GiB |
| Qwen3.5-9B-GLM5.1-Distill-v1 | vision + language | Q8_0 | 9.7B | 17.56 GiB | 4.76 GiB |
| Qwable-9B-Claude-Fable-5 | vision + language | F16 | 9.4B | 18.53 GiB | 3.79 GiB |
| Qwen3-VL-32B-Instruct | vision + language | Q3_K_S | 33.4B | 22.29 GiB | 0.03 GiB |
| Qwen3.5-9B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING | vision + language | BF16 | 9.4B | 18.53 GiB | 3.79 GiB |
| Qwen3.6-35B-A3BMoE | vision + language | Q4_K_M | 36.0B | 21.65 GiB | 0.67 GiB |
| gemma-4-12b-it-uncensored | vision + language | Q8_0 | 12.0B | 15.11 GiB | 7.21 GiB |
| gemma-4-26B-A4B-it-qat-q4_0-unquantized-uncensored-hereticMoE | vision + language | NVFP4 | 25.8B | 18.79 GiB | 3.53 GiB |
| Holo-3.1-9B | vision + language | F16 | 9.4B | 18.53 GiB | 3.79 GiB |
| MiniCPM-V-4_5 | vision + language | Q8_0 | 8.7B | 21.55 GiB | 0.77 GiB |
| MiniCPM-V-4.6 | vision + language | BF16 | 1.3B | 2.57 GiB | 19.75 GiB |
| Ornith-Agents-A1-3.6-35B-A3B-dare_tiesMoE | vision + language | Q4_K_M | 34.7B | 21.98 GiB | 0.34 GiB |
| Qwen3.5-9B-ultra-uncensored-heretic | vision + language | F16 | 9.4B | 18.53 GiB | 3.79 GiB |
| Qwen3-VL-8B-Thinking | vision + language | BF16 | 8.8B | 20.60 GiB | 1.72 GiB |
| Jan-v2-VL-high | vision + language | F16 | 8.8B | 20.60 GiB | 1.72 GiB |
| Huihui-gemma-4-26B-A4B-it-abliteratedMoE | vision + language | UD-Q5_K_M | 26.5B | 22.03 GiB | 0.29 GiB |
| LocateAnything-3B | vision + language | F16 | 3.8B | 10.46 GiB | 11.86 GiB |
| OvisOCR2 | vision + language | F32 | 853M | 3.97 GiB | 18.35 GiB |
This page models a generic 24GB accelerator, so it answers what fits rather than how fast it runs. For tokens per second you need a specific card — pick one from hardware, where bandwidth is known.