Unverified50% confidenceBenchmarkExact time
OpenAI's tests show that for 400-token texts, about 92% are detected when unedited, the detection rate drops to 66% after replacing 10% of words with synonyms, and only 17% after replacing 25% of words.
1
Sources
50%
Confidence
Long-term
Relevance
10/16/2026
First Seen
Sources
Related Entities
Related Claims
Unverified包含印地语注释的代码Token消耗可能比纯英文代码高出20-40%68% similarUnverified根据 Simon Willison 实测，新分词器导致英文文本 Token 增幅 +42%，西班牙文 +33%，Python 代码 +27%，中文基本无变化67% similarUnverifiedThe detection reliability of deeply rewritten content, non-English text, and short text is generally lower.67% similarUnverifiedThe detection rate of steganographic watermarks drops significantly after text is processed by paraphrase rewriting tools.67% similarUnverifiedFor English prose and OpenAI's current tokenizer, 1 token ≈ 4 characters, 1 token ≈ ¾ of a word, and a page of dense A4 text is about 600–800 tokens.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1042333API
curl https://kongchang.com/api/v1/knowledge/claims/1042333MCP
get_claim(id=1042333)