Unverified50% confidenceBenchmarkExact time
On the 0-100 "affected prose" scale, human text scores 6.65, Opus 5 scores 16.75, and Opus 5.5 drops to 10.57.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/16/2026
First Seen
Valid until: 1/14/2027
Sources
Related Entities
Related Claims
UnverifiedOpus 5.5 uses 0.015 em dashes per thousand words, a 99% decrease compared to Opus 5's 2.92.68% similarUnverifiedInter-annotator agreement among different human annotators in open-ended text quality assessment is typically only 0.6–0.8.68% similarUnverifiedAblation analysis shows that fine-tuning on the segment paradigm yields a +5.29-point improvement on the test set, while replacing the first 4,500 transcript words with 2,000 retrieved words yields only a +1.55-point improvement.65% similarUnverifiedAmong the five evaluators the authors tested, if the answer key had minor errors, the score was only 0.21 — lower than giving no key at all (0.46) and lower than giving a completely irrelevant key (0.50); while a human-verified model-generated answer scored 0.61, close to the 0.69 of a human-written one.65% similarUnverifiedFormat effect: with the same content changed to Markdown format, 4 of 5 evaluators preferred Markdown over plain text at rates of 73% to 97%, while two human annotators had only a 57% preference rate on 30 sample pairs.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1041458API
curl https://kongchang.com/api/v1/knowledge/claims/1041458MCP
get_claim(id=1041458)