Compare Local LLMs Side by Side

Pick 2–4 models and see exactly how much VRAM, disk space, and estimated tokens/sec each requires on your GPU, at every quantization level (Q2_K through F16).

What this tool shows

VRAM requiredModel weights + KV cache + runtime overhead at each quant level
Disk sizeGGUF file size on disk for the selected quantization
Est. tok/sBandwidth-roofline estimate on your selected GPU (not a benchmark)
Fit verdictGreat / Tight / Offload / Won't run — based on your VRAM
Ollama commandCopy-ready pull command for the recommended quantization

Estimates use the formula: tok/s ≈ GPU_bandwidth × 0.70 / active_weight_GB. Numbers are modelled estimates — see "How we estimate" in the tool for the full formula.