Unverified50% confidenceBenchmarkExact time
The LLM-oriented Markdown form saves 98% of tokens compared with raw HTML at the median and 63% compared with full-page-to-text conversion.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/13/2026
First Seen
Valid until: 1/11/2027
Sources
Related Entities
Related Claims
UnverifiedA typical HTML document may consume 2–5 times more tokens compared to equivalent Markdown content72% similarUnverifiedFeeding raw HTML directly to an LLM both consumes a large number of tokens and exacerbates hallucinations, making it necessary to strip boilerplate web content down to clean Markdown.68% similarUnverifiedHTML输出比Markdown消耗明显更多的Token，一个简单的加粗文字在Markdown中需要6个额外字符，而在HTML中需要17个额外字符68% similarUnverifiedThe raw HTML of the 20 pages totaled more than 3.71 million tokens, full-page-to-text conversion was about 243,000 tokens, main-content-only Markdown was about 110,000 tokens, and LLM Markdown was about 95,000 tokens.67% similarUnverifiedWhen raw HTML or basic BeautifulSoup loaders are fed directly to large language models, 70%–80% of context tokens are often consumed by irrelevant content such as navigation bars, cookie consent pop-ups, and advertisements.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1021866API
curl https://kongchang.com/api/v1/knowledge/claims/1021866MCP
get_claim(id=1021866)