ss-m3ultra-256-qwen35-122b-mlx-mxfp4 decode
57.0 tok/s (avg)
Fields
id
benchmarks/ss-m3ultra-256-qwen35-122b-mlx-mxfp4-tgMetric
decode_tok_s
Value
57
Unit
tok/s (avg)
Model
Qwen3.5-122B-A10B (MLX MXFP4)
Methodology
COMMUNITY-MEASURED row in siliconscore.com/benchmarks.csv (432-row machine-level export). Runtime MLX on ss-m3ultra-256-qwen35-122b-mlx-mxfp4. avg_tok_s is batch-1 decode-equivalent generation rate; prompt_eval_tok_s is the prefill column when present. context_tokens=unspecified. Row's upstream source_url=llmcheck; verified=false. Per-request batch-1, NOT aggregate. NOTE: siliconscore aggregates first-party lab runs + community reports (LocalScore, Reddit, llmcheck, sharpai); methodology per-row varies and is recorded as best-known.
Measured by
third_party
Reproducible
yes
notes
siliconscore row; lower confidence for community/unverified rows.
Notes
siliconscore row; lower confidence for community/unverified rows.