FIND YOUR LOCAL MODEL
Best local LLMs for your memory and task
Start with a memory budget, then compare one benchmark at a time. Recommendations update from the same 129-model catalog as the main chart.
Choose your RAM or VRAM
Each guide separates dedicated VRAM from system RAM and unified memory. Counts below use the system-memory planning reserve.
4 GB28 weight-budget candidates6 GB52 weight-budget candidates8 GB64 weight-budget candidates12 GB82 weight-budget candidates16 GB84 weight-budget candidates24 GB95 weight-budget candidates32 GB124 weight-budget candidates48 GB126 weight-budget candidates64 / 128 GB127 weight-budget candidates
64 GB and 128 GB share a guide that compares their weight budgets and available space. It shows additional 128 GB candidates when the catalog contains them.
Choose a comparison
Best local LLMs for codingCompare coding benchmark scores, weight sizes and Pareto trade-offs. LiveCodeBench v6 and SWE-bench Verified stay separate.Best local LLMs for reasoningCompare science and competition-math results separately, with exact quantization sizes and source-linked test settings.Best small local LLMsCompare models with at most 4 billion total parameters on knowledge, reasoning and coding tests. Small means total parameters, not active parameters.Local LLMs by memory footprintCompare exact weight bytes, quantization and total parameters. Weight footprint is a lower bound, not total runtime memory.Local LLM benchmark efficiencyCompare benchmark points per GB alongside Pareto trade-offs. A high score-to-size ratio alone does not select the best model.