Expired50% confidenceFactExact time
The deep-reading step downloads the PDFs of the 5–10 highest-scoring papers and extracts their full text, with a single paper's input reaching 40,000–70,000 tokens, requiring a model that supports long context.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/12/2026
First Seen
Valid until: 10/10/2026(expired)
Sources
Local AI Agent Reads arXiv So You Don't Have To: Only the Papers That Actually Matter
redditr/ollama7/10/2026
Related Claims
Unverified一篇标准学术论文全文通常包含 5,000 至 14,000 个 tokens72% similarUnverifiedOn older models such as Opus, the exact read rate at 5×8-pixel density was only about 10%; to reach 95%, the font had to be scaled up to 10×16, leaving room for only 7,000 characters per page.69% similarUnverified一张1568×728的生产页约需1400~1500个视觉Token，而同样内容作为高密度文本可能消耗1.5万到3万个Token69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/491237API
curl https://kongchang.com/api/v1/knowledge/claims/491237MCP
get_claim(id=491237)