LLMCheck: Apple M5 Max (128GB), MLX, Qwen3.6-35B-A3B Q4_K_M decode
55.0 tok/s (decode, per-request)
Fields
id
benchmarks/llmcheck-m5-max-qwen36-35b-a3b-mlx-decodeMetric
decode_tok_s
Value
55
Unit
tok/s (decode, per-request)
Model
Qwen3.6-35B-A3B
Methodology
THIRD-PARTY MEASURED (LLMCheck open dataset). M5 Max, MLX, Qwen3.6-35B-A3B Q4_K_M, TTFT 0.6 s, per-request decode.
Measured by
third_party
Reproducible
yes
notes
The repo's target-class model on the target-class chip. Ollama run of the same model/quant gives 48 tok/s — the MLX-vs-Ollama spread on an MoE is ~13%.
Notes
The repo's target-class model on the target-class chip. Ollama run of the same model/quant gives 48 tok/s — the MLX-vs-Ollama spread on an MoE is ~13%.