German Language LLM Index
a PeerBench project

Individual experiment result

MuSR (DE)

DeepSeek V4 Flash DeepSeek

Accuracy reasoning off fp8
83.5%
Accuracy

Run details

Date
2026-06-10
Test cases
564
Median latency
10.18s
Total cost
$0.26
Avg. prompt tokens
1,659.8
Avg. answer tokens
798.9
Avg. reasoning tokens
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
object placements 75.0%
team allocation 84.8%

Provenance

Run ID
2026-06-10T11-50-02__openrouter__deepseek__deepseek-v4-flash__musr_de