IMMUTABLE RUN / 2026-08-28
Documented local resultFlash-Next Q3_K_XL
Deterministic agent benchmark · Unraid 128 GB + RTX 3060
9/9tasks passed
14.5decode tok/s
CONFIGURATION
- Runtime
- llama.cpp local-build
- Quant
- Q3_K_XL
- Context
- 32,768 tokens
- Concurrency
- 1
PROVENANCE
Transcribed from Unraid Qwen3.8 notes, dated 2026-08-28.
- Transcribed from a local note
- Raw run output is unavailable
CAVEATS
- CPU-offloaded MoE
- Latency is materially higher than Qwen3.6
- Benchmark version
- 1
- Task-set hash
sha256:documented-agent-v1-nine-tasks- Run ID
flash-next-q3kxl-agent-2026-08-28