Archive · vps-81 historical telemetry · local-mac/2026-04-11-mac-qwen-coder-ladder-overview.html. Originally rendered 2026-04-11. Re-hosted from MyServers on 2026-05-06. Methodology and harness conventions may differ from what we use today; see /methodology.html for current standards. ← back to all benchmarks

Mac Qwen Coder Ladder

This page compares the local Apple-silicon Mac runs for qwen2.5-coder:0.5b, qwen2.5-coder:1.5b, and qwen2.5-coder:3b under the shared benchmark stack.

Qwen 0.5B 314.5s

Shared local-Mac total for 5Q, 20Q, and 10Q.

Hello: 1.8s · 20Q avg: 4.4s · 10Q avg: 8.4s

Qwen 1.5B 769.4s

Shared local-Mac total for 5Q, 20Q, and 10Q.

Hello: 4.8s · 20Q avg: 11.4s · 10Q avg: 24.5s

Qwen 3B 914.5s

Shared local-Mac total for 5Q, 20Q, and 10Q.

Hello: 5.8s · 20Q avg: 13.3s · 10Q avg: 31.1s

How To Read This

This page compares only the local Apple-silicon Mac runs under the same one-model Ollama shape: one loaded model, one parallel slot, 4096 context, and the same benchmark packets. That makes it the cleanest view of how model size changes behavior on this Mac without the VPS in the loop.

Five-Question Packet

Metric Qwen 0.5B Qwen 1.5B Qwen 3B
Total wall time88.0s68.6s110.1s
Average question time17.6s13.7s22.0s
Average throughput45.69 tok/s21.17 tok/s12.86 tok/s
Average marker hit80%83%90%
Format passes3/52/53/5
Strict passes2/52/53/5

Python 20Q

Metric Qwen 0.5B Qwen 1.5B Qwen 3B
Total wall time122.2s378.6s397.1s
Primary avg duration4.4s11.4s13.3s
Follow-up avg duration1.7s7.5s6.5s
Primary avg throughput42.18 tok/s16.89 tok/s14.20 tok/s
Primary avg marker hit78%88%85%
Usable primary answers20/2020/2020/20

Real-Context 10Q

Metric Qwen 0.5B Qwen 1.5B Qwen 3B
Total wall time104.3s322.2s407.3s
Primary avg duration8.4s24.5s31.1s
Follow-up avg duration2.0s7.7s9.6s
Primary avg throughput38.14 tok/s17.82 tok/s14.01 tok/s
Primary avg marker hit55%69%78%
Usable primary answers10/1010/1010/10

Drill-Down Reports

Every link below goes to a standalone archived report page.