Benchmark result
qwen3.8-27b-splash on Apple M4 Max — 69.8 tok/s
Measured with LM Studio on October 3, 2026.
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
89th fastest of 97 runs of this model on Apple M4 Max · median 73.6 tok/s.
sort
- 179.1tok/s88.3LM Studio · 4bit
- 278.1tok/s83.7LM Studio · 4bit
- 377.7tok/s80.2LM Studio · 4bit
- 477.5tok/s82.1LM Studio · 4bit
- 577.4tok/s85.6LM Studio · 4bit
- 677.2tok/s82.7LM Studio · 4bit
- 776.9tok/s83.5LM Studio · 4bit
- 876.8tok/s84.4LM Studio · 4bit
- 976.7tok/s81.0LM Studio · 4bit
- 1076.7tok/s85.5LM Studio · 4bit
- 1176.6tok/s87.3LM Studio · 4bit
- 1276.5tok/s80.2LM Studio · 4bit
- 1376.4tok/s86.2LM Studio · 4bit
- 1476.4tok/s86.2LM Studio · 4bit
- 1576.3tok/s81.3LM Studio · 4bit
- 1676.3tok/s87.1LM Studio · 4bit
- 1776.2tok/s86.2LM Studio · 4bit
- 1876.0tok/s84.6LM Studio · 4bit
- 1976.0tok/s87.1LM Studio · 4bit
- 2076.0tok/s78.1LM Studio · 4bit
- 2175.9tok/s81.1LM Studio · 4bit
- 2275.6tok/s79.0LM Studio · 4bit
- 2375.4tok/s84.4LM Studio · 4bit
- 2475.3tok/s84.3LM Studio · 4bit
- 2575.3tok/s79.4LM Studio · 4bit
- 2675.1tok/s83.0LM Studio · 4bit
- 2775.0tok/s82.5LM Studio · 4bit
- 2875.0tok/s83.5LM Studio · 4bit
- 2974.9tok/s82.8LM Studio · 4bit
- 3074.8tok/s86.5LM Studio · 4bit
- 3174.7tok/s82.5LM Studio · 4bit
- 3274.7tok/s81.6LM Studio · 4bit
- 3374.6tok/s86.3LM Studio · 4bit
- 3474.6tok/s83.7LM Studio · 4bit
- 3574.5tok/s82.3LM Studio · 4bit
- 3674.5tok/s82.1LM Studio · 4bit
- 3774.4tok/s84.2LM Studio · 4bit
- 3874.3tok/s82.7LM Studio · 4bit
- 3974.3tok/s78.4LM Studio · 4bit
- 4074.2tok/s79.9LM Studio · 4bit
- 4174.0tok/s85.3LM Studio · 4bit
- 4274.0tok/s79.9LM Studio · 4bit
- 4374.0tok/s85.9LM Studio · 4bit
- 4473.9tok/s83.6LM Studio · 4bit
- 4573.9tok/s82.1LM Studio · 4bit
- 4673.9tok/s81.8LM Studio · 4bit
- 4773.8tok/s82.9LM Studio · 4bit
- 4873.7tok/s82.3LM Studio · 4bit
- 4973.6tok/s85.4LM Studio · 4bit
- 5073.5tok/s82.2LM Studio · 4bit
- 5173.4tok/s79.8LM Studio · 4bit
- 5273.4tok/s83.8LM Studio · 4bit
- 5373.3tok/s77.9LM Studio · 4bit
- 5473.3tok/s82.4LM Studio · 4bit
- 5573.2tok/s87.8LM Studio · 4bit
- 5673.0tok/s74.9LM Studio · 4bit
- 5773.0tok/s83.1LM Studio · 4bit
- 5872.9tok/s76.9LM Studio · 4bit
- 5972.9tok/s83.4LM Studio · 4bit
- 6072.9tok/s81.2LM Studio · 4bit
- 6172.7tok/s81.9LM Studio · 4bit
- 6272.6tok/s84.6LM Studio · 4bit
- 6372.6tok/s83.5LM Studio · 4bit
- 6472.6tok/s75.5LM Studio · 4bit
- 6572.4tok/s81.4LM Studio · 4bit
- 6672.4tok/s78.8LM Studio · 4bit
- 6772.4tok/s83.1LM Studio · 4bit
- 6872.2tok/s83.2LM Studio · 4bit
- 6972.2tok/s83.5LM Studio · 4bit
- 7072.2tok/s79.0LM Studio · 4bit
- 7172.2tok/s80.9LM Studio · 4bit
- 7272.1tok/s83.5LM Studio · 4bit
- 7372.1tok/s84.8LM Studio · 4bit
- 7472.0tok/s86.0LM Studio · 4bit
- 7571.8tok/s80.3LM Studio · 4bit
- 7671.7tok/s86.0LM Studio · 4bit
- 7771.7tok/s80.3LM Studio · 4bit
- 7871.6tok/s84.9LM Studio · 4bit
- 7971.6tok/s82.7LM Studio · 4bit
- 8071.4tok/s80.9LM Studio · 4bit
- 8171.3tok/s76.5LM Studio · 4bit
- 8271.2tok/s79.7LM Studio · 4bit
- 8371.1tok/s81.7LM Studio · 4bit
- 8470.7tok/s83.4LM Studio · 4bit
- 8570.7tok/s78.5LM Studio · 4bit
- 8670.5tok/s84.7LM Studio · 4bit
- 8770.4tok/s84.2LM Studio · 4bit
- 8870.1tok/s84.7LM Studio · 4bit
- 8969.8tok/s85.2LM Studio · 4bitthis run
- 9069.7tok/s83.4LM Studio · 4bit
- 9169.6tok/s84.3LM Studio · 4bit
- 9269.6tok/s83.0LM Studio · 4bit
- 9369.5tok/s80.9LM Studio · 4bit
- 9469.2tok/s83.6LM Studio · 4bit
- 9567.8tok/s84.7LM Studio · 4bit
- 9666.7tok/s83.0LM Studio · 4bit
- 9766.0tok/s83.2LM Studio · 4bit
Reproduce this run
toolLM Studio
modelqwen3.8-27b-splash (4bit)
context65,536
clientv0.4.87+102
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.