Best local AI models for 6GB VRAM
Ranked by what actually fits at 32K context, computed from real file bytes.
A 6GB card gives you about 5.58 GiB to work with after driver overhead. 647 indexed models fit at 32K context — the largest being granite-20b-code-instruct-8k at 20.1B parameters in IQ1_M.
From the file· fit from summed bytesFrom the file· KV per layer
allvision + language 67speech recognition 36text generation 502speech synthesis 18embeddings 21video generation 3
Fits in 6GB at 32K context
largest quantization that fits, per model
| Model | Modality | Best quant | Params○ | Total◐ | Headroom◐ |
|---|---|---|---|---|---|
| Qwen3.5-9B | vision + language | UD-IQ2_M | 9.7B | 5.53 GiB | 0.05 GiB |
| nemotron-3.5-asr-streaming-0.6b | speech recognition | F32 | 638M | 3.22 GiB | 2.36 GiB |
| Qwen3.5-4B | vision + language | Q6_K_L | 4.7B | 5.50 GiB | 0.08 GiB |
| gemma-4-E4B-it | text generation | Q3_K_S | 8.0B | 4.92 GiB | 0.66 GiB |
| Qwen3.5-0.8B | vision + language | BF16 | 873M | 2.60 GiB | 2.98 GiB |
| Llama-3.2-1B-Instruct | text generation | F16 | 1.2B | 4.11 GiB | 1.47 GiB |
| gemma-4-E2B-it | vision + language | Q6_K_L | 5.1B | 5.29 GiB | 0.29 GiB |
| parakeet-tdt-0.6b-v3 | speech recognition | F32 | 627M | 3.18 GiB | 2.40 GiB |
| ced-base | text generation | F32 | 86M | 1.16 GiB | 4.42 GiB |
| Ace-Step1.5 | speech synthesis | F32 | 160M | 2.71 GiB | 2.87 GiB |
| Qwen2.5-7B-Instruct | text generation | Q2_K | 7.6B | 5.41 GiB | 0.17 GiB |
| Qwen3-TTS-12Hz-0.6B-Base | speech synthesis | Q4_K_M | 915M | 5.57 GiB | 0.01 GiB |
| Qwythos-9B-v2 | vision + language | I1-Q2_K | 9.7B | 5.48 GiB | 0.10 GiB |
| UI-TARS-1.5-7B | text generation | Q2_K | 8.3B | 5.41 GiB | 0.17 GiB |
| gemma-3-1b-it | text generation | F16 | 1000M | 2.81 GiB | 2.77 GiB |
| gemma-4-E2B-it-qat-q4_0-unquantized | text generation | Q4_K | 5.1B | 4.22 GiB | 1.36 GiB |
| Qwen3-1.7B | text generation | Q4_1 | 2.0B | 5.54 GiB | 0.04 GiB |
| whisper-medium | speech recognition | F32 | 764M | 3.69 GiB | 1.89 GiB |
| embeddinggemma-300m | text generation | F32 | 303M | 2.05 GiB | 3.53 GiB |
| Llama-3.2-3B-Instruct | text generation | Q2_K | 3.2B | 5.58 GiB | 0.00 GiB |
| Qwen3-0.6B | text generation | Q8_0 | 752M | 5.02 GiB | 0.56 GiB |
| Wan2.1-T2V-1.3B | text generation | Q5_K_M | 1.4B | 3.44 GiB | 2.14 GiB |
| LFM2.5-1.2B-Instruct | text generation | BF16 | 1.2B | 3.37 GiB | 2.21 GiB |
| Qwen2.5-Coder-7B-Instruct | text generation | Q2_K | 7.6B | 5.41 GiB | 0.17 GiB |
| Qwen2.5-1.5B-Instruct | text generation | F16 | 1.5B | 4.56 GiB | 1.02 GiB |
| gemma-3-4b-it | text generation | Q8_0 | 4.3B | 5.46 GiB | 0.12 GiB |
| whisper-large-v3 | speech recognition | F16 | 1.5B | 3.74 GiB | 1.84 GiB |
| MiniCPM5-1B-Claude-Opus-Fable5-Thinking | text generation | F16 | 1.1B | 3.55 GiB | 2.03 GiB |
| Qwen2.5-VL-7B-Instruct | vision + language | UD-IQ3_XXS | 8.3B | 5.55 GiB | 0.03 GiB |
| Qwen3-VL-2B-Instruct | vision + language | Q5_K_L | 2.1B | 5.54 GiB | 0.04 GiB |
| Ornith-1.0-9B | text generation | IQ2_M | 9.2B | 5.44 GiB | 0.14 GiB |
| jina-embeddings-v5-text-small | embeddings | F16 | 596M | 5.39 GiB | 0.19 GiB |
| whisper-large-v3-turbo | speech recognition | F16 | 809M | 2.36 GiB | 3.22 GiB |
| granite-4.1-3b | text generation | Q5_K_M | 3.4B | 5.57 GiB | 0.01 GiB |
| Qwen3.5-2B | vision + language | BF16 | 2.3B | 4.80 GiB | 0.78 GiB |
| Qwen2.5-3B-Instruct | text generation | Q8_0 | 3.1B | 5.30 GiB | 0.28 GiB |
| gemma-2-2b-it | text generation | Q8_0 | 2.6B | 5.26 GiB | 0.32 GiB |
| MiniCPM5-1B-Claude-Opus-Fable5-V2-Thinking | text generation | F16 | 1.1B | 3.55 GiB | 2.03 GiB |
| Qwen2.5-0.5B-Instruct | text generation | F16 | 494M | 2.08 GiB | 3.50 GiB |
| embeddinggemma-300m-qat-q8_0-unquantized | embeddings | Q8_0 | 303M | 1.22 GiB | 4.36 GiB |
| Qwen3-ASR-1.7B | speech recognition | F16 | 2.3B | 5.23 GiB | 0.35 GiB |
| TinyLlama-1.1B-Chat-v1.0 | text generation | F16 | 1.1B | 3.53 GiB | 2.05 GiB |
| Qwen3-ASR-0.6B | speech recognition | F16 | 938M | 2.32 GiB | 3.26 GiB |
| SmolLM2-135M-Instruct | text generation | F16 | 135M | 1.72 GiB | 3.86 GiB |
| DeepSeek-R1-Distill-Qwen-7B | text generation | Q2_K | 7.6B | 5.41 GiB | 0.17 GiB |
| GLM-4.6V-Flash | text generation | UD-IQ2_XXS | 10.3B | 5.36 GiB | 0.22 GiB |
| Qwen2.5-Coder-3B-Instruct | text generation | Q8_0 | 3.1B | 5.30 GiB | 0.28 GiB |
| KaLM-embedding-multilingual-mini-instruct-v2.5 | embeddings | Q8_0 | 494M | 1.65 GiB | 3.93 GiB |
| nomic-embed-text-v1.5 | embeddings | F32 | 137M | 2.40 GiB | 3.18 GiB |
| Qwen2.5-Coder-1.5B-Instruct | text generation | F16 | 1.5B | 4.56 GiB | 1.02 GiB |
| GigaAM-v3 | speech recognition | F32 | 223M | 1.67 GiB | 3.91 GiB |
| gemma-3-270m-it | text generation | F16 | 268M | 1.38 GiB | 4.20 GiB |
| umt5-xxl | text generation | Q6_K | 5.7B | 5.20 GiB | 0.38 GiB |
| DeepSeek-R1-Distill-Qwen-1.5B | text generation | BF16 | 1.8B | 4.99 GiB | 0.59 GiB |
| LFM2.5-8B-A1BMoE | text generation | UD-IQ4_NL | 8.5B | 5.23 GiB | 0.35 GiB |
| jina-embeddings-v5-text-nano | embeddings | F16 | 212M | 2.30 GiB | 3.28 GiB |
| SmolVLM-500M-Instruct | text generation | F16 | 507M | 2.78 GiB | 2.80 GiB |
| tinygemma3_cifar | text generation | Q8_0 | 39M | 0.93 GiB | 4.65 GiB |
| LFM2.5-VL-1.6B | vision + language | BF16 | 1.6B | 3.37 GiB | 2.21 GiB |
| Ministral-3-3B-Reasoning-2512 | text generation | Q2_K | 4.3B | 5.42 GiB | 0.16 GiB |
This page models a generic 6GB 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.