Local LM Speed Lab

Decode throughput of local models on home hardware

Benchmark runs

–

 

Models tested

–

 

Best prose speed

–

 

Best code speed

–

 

Speed per machine

Best decode throughput per machine, prose vs code

Prose vs code

Each dot is one run · tok/s

Recipe leaderboard

Every run ranked by prose decode speed

All runs

Click a column header to sort

# Model Machine Memory Recipe Prose tok/s Code tok/s