gemini-4POST
自建 OCR benchmark：Gemini 4 Argon 不如 Gemini 3.0 Flash
用户用一批难处理的 PDF 转 Markdown 测试，认为 Gemini 4 Argon 在 OCR 上不如 Gemini 3.0 Flash，更倾向编造内容而非如实提取。
Johannes Schmidt
@spaceMonster5 天前收录原推发布于 2026.10.26
Johannes Schmidt @spaceMonster原推发布于 2026.10.26
@theo Gemini 4 Argon is worse than Gemini 3.0 Flash in OCR tasks in my own benchmark (i should publish that soon). I have a parkour of tricky PDFs and it's task is to convert them to markdown Gemini 4 is more comfortable to lie about content than to extract the messy content as is
x.com/spacemonster/status/2114484817937027439采集于 2026.10.26 10:01