German Language LLM Index
a PeerBench project

Individual experiment result

SB10K (DE)

Qwen3.8 27B (FP8) Alibaba

Matthews correlation coefficient reasoning off fp8
62.7%
Matthews correlation coefficient

Run details

Date
2026-08-15
Test cases
1,024
Median latency
2.27s
Total cost
Avg. prompt tokens
819.9
Avg. answer tokens
2.7
Avg. reasoning tokens
Quantization
fp8

About this benchmark

Native German Twitter sentiment (positive/neutral/negative), generate_until variant; primary metric MCC.

3-class sentiment native · Native German Source ↗

Provenance

Canonical model
qwen/qwen3.8-27b-fp8
Run ID
2026-08-15T08-48-21__openai__Qwen__Qwen3.8-27B-FP8__sb10k_de