Key Info

A benchmark table shared by Burke Holland reports warm tokens-per-second (TPS) and cold time-to-first-token (TTFT) for several Ollama models running locally on a laptop with 128GB of unified RAM.

Highlights

  • gemma4:e4b: 122.8 warm TPS, 8.65s cold TTFT
  • qwen3-coder:30b: 90 warm TPS, 10.25s cold TTFT
  • gemma4:26b-a4b: 86.9 warm TPS, 13.03s cold TTFT
  • gpt-oss:20b: 60.7 warm TPS, 9.93s cold TTFT
  • qwen3.8:latest: 25.3 warm TPS, 14.84s cold TTFT
  • The tester described the local performance as feeling quick for a laptop.