German Language LLM Index
a PeerBench project

Individual experiment result

MuSR (DE)

Claude Fable 5 Anthropic

Accuracy reasoning on / locked provider-internal
89.5%
Accuracy

Run details

Date
2026-07-05
Test cases
564
Median latency
20.28s
Total cost
$54.71
Avg. prompt tokens
3,008.8
Avg. answer tokens
1,338.2
Avg. reasoning tokens
Quantization
provider-internal

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 92.8%
object placements 81.3%
team allocation 88.4%

Provenance

Run ID
2026-07-05T22-33-14__openai__cc__claude-fable-5__musr_de