German Language LLM Index
a PeerBench project

Individual experiment result

MuSR (DE)

DeepSeek V4 Flash 0423 DeepSeek

Accuracy reasoning off fp4
85.3%
Accuracy

Run details

Date
2026-08-11
Test cases
564
Median latency
16.41s
Total cost
$0.16
Avg. prompt tokens
1,659.8
Avg. answer tokens
801
Avg. reasoning tokens
Quantization
fp4

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 87.6%
object placements 71.9%
team allocation 86.4%

Provenance

Run ID
2026-08-11T12-39-19__openrouter__deepseek__deepseek-v4-flash__musr_de