ModelDock
3,571.4
132.5
476.9
MODELDOCK / LOCAL ECONOMICS
Best in Speed
A combined local-hardware speed ranking for Qwen3.8 27B: 75% Reading and 25% Writing, weighted by the actual time each portion takes.
Hardware × model dataset
SORTED BYWEIGHTED TOKEN SPEEDHigher is better
LIVE BENCHMARK SNAPSHOT
Hardware configurations, ranked.
6,288
112.5
427.1
7,756
109.3
419.3
3,250
113.3
410.1
05
Mac Studio M5 Ultra 256 GBEstimate
~1,075
~91
290.3
06
Mac Studio M5 Ultra 96 GBEstimate
~1,075
~91
290.3
07
2,692.3
68.8
255.6
1,458.3
69.8
244
09
2× RTX 4080Estimate
~1,700
~58
210.5
723
65.9
207
745
59.2
191.2
12
Mac Studio M5 Max 128 GBEstimate
~745
~59.2
191.2
13
Mac Studio M5 Max 64 GBEstimate
~745
~59.2
191.2
14
705
55.9
180.6
1,225.5
44
159
877.8
43.9
152.7
424.8
49.3
146.4
462.5
44.3
137.5
19
Mac Studio M4 Max 64 GBEstimate
~274
~53.3
134.6
20
Mac mini M5 Pro 64 GBEstimate
~373
~29.6
95.6
WHAT IS INCLUDED
Hardware configurations that fit the ordinary Qwen3.8 27B path and disclose reading, writing, power, and acquisition inputs.
WHY THIS FORMULA
It adds time, not rates: a slow write phase cannot disappear behind a fast prefill rate.
EVIDENCE
Open the Benchmark workbench to inspect source medians, conditions, economics assumptions, and price references.