Unverified50% confidenceBenchmarkExact time
The benchmark is built on the Kaggle Benchmarks platform; each model is required to answer 287 scored questions spanning five dimensions: numbers, verbs, translation, cloze, and error correction.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/11/2026
First Seen
Valid until: 1/9/2027
Sources
Related Entities
Related Claims
UnverifiedThe key metric, the headline score, refers to the proportion of sentences fully and correctly restored; if even one tone mark is wrong the entire sentence is counted as incorrect, and the test is built on the kaggle-benchmarks library.71% similarUnverifiedThe researchers constructed a benchmark test set called ALIGNMENTFORECASTBENCH, covering more than 5,000 prediction questions, 17 target models, 32 datasets, and 16 failure modes.69% similarUnverifiedIndependent evaluator Anass Kartit tested all decision models available on OpenRouter using 82 manually annotated questions, and additionally tested 4 models locally on a Mac.68% similarUnverifiedA developer built a benchmark called Unlearning Security Audit in the Kaggle Benchmarking Challenge to test whether large language models can detect disguised unlearning code.68% similarUnverifiedEach test point is output in a unified structured format containing the module it belongs to, title, status, steps, expected results, priority, and test design method used.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/999197API
curl https://kongchang.com/api/v1/knowledge/claims/999197MCP
get_claim(id=999197)