Model Detail
Qwen 3.6 27B NVFP4
Best current 27B result of the three, but still too soft to trust as a default local operator model.
Best of the weak pass
- Rank
- #11 of 13
- Source
- Enterprise oMLX
- Run date
- Avg latency
- 17.44s
01 · Benchmark score 40/100
02 · Average latency 17.44s
03 · Role Best of the weak pass
Per-suite breakdown
Scorecard
Operator board #13
40/100 operator score
Messaging Quick execution pack
Messaging Tool Planning v2
External canon First internal 27B oMLX pass
Public benchmark references
Cost Enterprise local
Best current 27B result of the three, but still too soft to trust as a default local operator model.
Strengths
- Tied top score in this 27B run
- Live on oMLX/OpenClaw
- Clean no-thinking API path
Weaknesses
- Only 40/100 on the quick pack
- Missed multiple operator-judgment tasks
- Needs rerun/tuning before routing decisions
Best current 27B result of the three, but still too soft to trust as a default local operator model.
Comparison
OperatorIndex Where Qwen 3.6 27B NVFP4 lands
- 01 Gemini 3.1 ProMessaging benchmark + external canon 100
- 01 Claude Sonnet 4.6Messaging benchmark + external canon 100
- 01 Gemini FlashMessaging benchmark canon 100
- 04 Claude Opus 4.6Operator Suite v2 95.3
- 05 GLM-5-TurboOperator Suite v2 95
- 06 Gemma 4 31BEnterprise Ollama 92
- 07 MiniMax M2.7Operator Suite v2 90.8
- 08 GPT-5.4External benchmark canon 88
- 09 Gemma 4 26BEnterprise Ollama 80
- 10 PrismML Bonsai 1.7BPrismML local benchmark 56
- 11 Qwen 3.6 27B NVFP4Enterprise oMLX 40
- 11 Qwen 3.6 27B MXFP4Enterprise oMLX 40
- 13 Qwen 3.6 27B 4bitEnterprise oMLX 20
Source artifacts
Raw machine-readable files for anyone who wants to dig deeper or run their own analysis.
- internal artifact output/benchmarks/2026-04-25-qwen36-27b-api/README.md
- internal artifact output/benchmarks/2026-04-25-qwen36-27b-api/results-raw.json
- internal artifact output/benchmarks/2026-04-25-qwen36-27b-api/results-scored.json