Benchmark result
qwen3.8-27b-splash on Apple M5 Max — 99.5 tok/s
Measured with LM Studio on October 14, 2026.
Support llm-bench.io
llm-bench.io is free. If it helped you, I'd appreciate your support to cover the infrastructure costs.
What the model built
Open full screen →The coding scenario asks for a playable game in a single HTML file. This is exactly what the model returned, unedited; the frame only adds a line that tells this page how tall it is.
How this run compares
2nd fastest of 13 runs of this model on Apple M5 Max · median 73.1 tok/s.
sort
- 1101.7tok/s80.8LM Studio · 4bit
- 299.5tok/s80.6LM Studio · 4bitthis run
- 376.6tok/s75.3LM Studio · 4bit
- 476.3tok/s84.7LM Studio · 4bit
- 575.3tok/s84.6LM Studio · 4bit
- 674.4tok/s84.3LM Studio · 4bit
- 773.1tok/s73.8LM Studio · 4bit
- 873.0tok/s85.0LM Studio · 4bit
- 972.7tok/s85.2LM Studio
- 1072.4tok/s74.5LM Studio · 4bit
- 1171.5tok/s86.0LM Studio · 4bit
- 1271.2tok/s83.4LM Studio · 4bit
- 1371.1tok/s81.6LM Studio
Reproduce this run
toolLM Studio
modelqwen3.8-27b-splash (4bit)
context131,072
clientv0.4.85+101
Set those in LM Studio, then:
llm-benchmark benchmark --model "qwen3.8-27b-splash" --tool "LM Studio"The client prompts for context length and thinking mode, and for the KV cache dtype on Unsloth Studio. Sampling is left at the model default — the values above are what the tool reported using, not overrides the client sent.