Unverified60% confidenceFactExact time
该模型基于V4 Flash架构，通过持续训练（continued training）叠加视觉模块获得图像理解能力
2
Sources
60%
Confidence
Long-term
Relevance
9/3/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedV4-Flash-Vision-Exp能够读取截图、解析图表，并结合工具调用完成多模态Agent任务78% similarUnverifiedV4-Flash-Vision-Exp在纯文本Agent能力上与V4-Flash基本持平73% similarUnverifiedDeepSeek将实验性多模态模型V4-Flash-Vision-EXP上线API平台，其文本能力与V4 Flash持平73% similarUnverifiedDINOv2是Meta AI于2023年发布的视觉基础模型，采用自监督学习范式在大规模无标注图像数据上进行预训练72% similarUnverified谷歌的PaLI、OpenAI的GPT-4V、Meta的LLaMA Vision均沿用了类似的图文对齐预训练策略71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/849870API
curl https://kongchang.com/api/v1/knowledge/claims/849870MCP
get_claim(id=849870)