16 GB is the sweet spot: 14B with room to spare, 20–27B with the right quantization. Below are the 29 models we track, with the highest-quality quantization that fits in 16 GB, memory needed and estimated speed (8K context, data updated 2026-09-27).
Calculate with your own settings →
Speed estimated on the RTX 5060 Ti 16GB (448 GB/s).
| Model | Best quant | Memory (8K) | Verdict | Speed |
|---|---|---|---|---|
| Qwen3.5 4B Qwen · Lightweight, image input | BF16 | 10.5GB | Comfortable | Usable (34 tok/s) |
| Gemma 4 E4B Gemma · Lightweight for phones/laptops | Q8_0 | 9.5GB | Comfortable | Usable (37 tok/s) |
| Qwen3 8B Qwen · The go-to 8B | Q8_0 | 10.8GB | Comfortable | Usable (36 tok/s) |
| Llama 3.1 8B Llama · Classic, many fine-tunes | Q8_0 | 10.5GB | Comfortable | Usable (37 tok/s) |
| Ministral 3 8B Mistral | Q8_0 | 11.5GB | Comfortable | Usable (33 tok/s) |
| DeepSeek-R1-0528-Qwen3-8B DeepSeek · Reasoning (thinking) model | Q8_0 | 10.8GB | Comfortable | Usable (36 tok/s) |
| llm-jp-4 8B thinking Japanese · Japanese-focused (Japan) | Q8_0 | 11.1GB | Comfortable | Usable (34 tok/s) |
| Qwen3-Swallow 8B Japanese · Japanese-enhanced (Institute of Science Tokyo) | Q8_0 | 10.8GB | Comfortable | Usable (36 tok/s) |
| Qwen3.5 9B Qwen · Image input | Q8_0 | 11.4GB | Comfortable | Usable (31 tok/s) |
| Gemma 4 12B Gemma · Mid-size, good multilingual | Q8_0 | 14.1GB | Comfortable | Usable (25 tok/s) |
| Ministral 3 14B Mistral | Q6_K | 13.8GB | Comfortable | Usable (27 tok/s) |
| Qwen3 14B Qwen | Q6_K | 14.5GB | Tight | Usable (26 tok/s) |
| gpt-oss-20b OpenAI · OpenAI open-weight model (MXFP4) | Q4_K_M (MXFP4) | 13.2GB | Comfortable | Fast (91 tok/s) |
| Devstral Small 2 24B Mistral · Coding | IQ4_XS | 15.5GB | Tight | Usable (24 tok/s) |
| Gemma 4 26B-A4B Gemma · Fast MoE | IQ4_XS | 15.5GB | Tight | Fast (87 tok/s) |
| Qwen3.8 27B Qwen · Latest generation (Aug 2026) | Q3_K_M | 15.2GB | Tight | Usable (23 tok/s) |
| Model | Best quant | Memory (8K) | Verdict | Speed |
|---|---|---|---|---|
| Qwen3-Coder 30B-A3B Qwen · Coding MoE | Q4_K_M | 20.4GB | Spills to RAM | Usable (32 tok/s) |
| GLM-4.7-Flash GLM · MoE, strong at long context | Q4_K_M | 20.4GB | Spills to RAM | Usable (35 tok/s) |
| Nemotron 3 Nano 30B-A3B NVIDIA · Mamba hybrid, light long context | Q4_K_M | 20.3GB | Spills to RAM | Usable (31 tok/s) |
| Gemma 4 31B Gemma · Largest Gemma 4 | Q4_K_M | 21.5GB | Spills to RAM | Very slow (4.8 tok/s) |
| Qwen3 32B Qwen | Q4_K_M | 23.1GB | Spills to RAM | Very slow (4.0 tok/s) |
| Qwen3-Swallow 32B Japanese · Japanese-enhanced (Institute of Science Tokyo) | Q4_K_M | 23.1GB | Spills to RAM | Very slow (4.0 tok/s) |
| llm-jp-4 33B thinking Japanese · Japanese-focused (Japan) | Q4_K_M | 23.4GB | Spills to RAM | Very slow (3.9 tok/s) |
| Qwen3.6 35B-A3B Qwen · Popular fast MoE | Q4_K_M | 23.1GB | Spills to RAM | Usable (28 tok/s) |