Qwen3.6-27B Quantized Versions · BFCL-v3 Evaluation Visualization

174 test samples Benchmark Scripts

Local OpenAI API @ 1235 EvalScope BFCL-v3 Agent / Function Calling temperature = 0

Filter Models
Top N Models
by OVERALL score
Select Models

Overview

Capability Comparison

Accuracy Radar Chart

Subset Accuracy Comparison

Performance Metrics

Speed: Prefill vs Decode

VRAM Usage & Completion Time

Full Data (click headers to sort)

Model OVERALL PM JS LPM irrelevance MTB MTLC VRAM (GB) Context (k) Completion Time

Note: MTB = Multi-Turn Base; PM = Parallel Multiple; LPM = Live Parallel Multiple; JS = JavaScript. MTB score is shown as "-" if not tested; OVERALL accuracy is based on the first 4 subsets.
Completion time refers to overall evaluation duration, not per-instance latency.