German Language LLM Index
a PeerBench project

Individual experiment result

MuSR (DE)

GLM-5.3 Flash Z.ai

Accuracy reasoning on / locked fp8
81.9%
Accuracy

Run details

Date
2026-08-30
Test cases
553
Median latency
19.89s
Total cost
$0.0038
Avg. prompt tokens
1,612.1
Avg. answer tokens
386.6
Avg. reasoning tokens
98.7
Quantization
fp8

About this benchmark

German MuSR v1.1 defect-filtered benchmark: 553 questions (11 truncated team-allocation translations excluded), generate_until cot+; primary metric accuracy.

2–5 option multiple choice translated · German translation Source ↗

Subject breakdown

Subject Score
murder mystery 81.6%
object placements 68.8%
team allocation 85.8%

Provenance

Canonical model
z-ai/glm-5.3-flash
Run ID
2026-08-30T16-17-19__openrouter__z-ai__glm-5.3-flash__musr_de