Expired50% confidenceFactExact time
In multimodal tool call configuration, History must be enabled for the vision model; otherwise, the vision model cannot access context information that has already occurred in the preceding conversation.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026(expired)
Sources
1FlowBase in Practice: Adding Vision Tools to DeepSeek V4 for Multimodal Capabilities
bilibili老文-taichuy6/21/2026
Related Claims
Unverified对于需要精确视觉推理的场景（文档理解、图表问答、机器人视觉、医学影像），单靠CLIP视觉编码器不够，建议在视觉编码器层面做增强而非仅依赖更大的语言模型68% similarUnverified该方法通过视觉示例而非文本描述来定义目标类别，与OWL-ViT、Grounding DINO等通过视觉-语言对齐实现开放词汇能力的方法不同67% similarUnverified视觉幻觉指模型在描述图像时生成与实际视觉内容不符的信息，源于语言模型倾向于用语言先验知识填补视觉信号不明确的空白67% similarUnverifiedACAI的语义记忆图谱主张将长期上下文映射为互联的知识图谱，形成用户→项目→技术栈的层级关联，以缓解上下文窗口腐化65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/43438API
curl https://kongchang.com/api/v1/knowledge/claims/43438MCP
get_claim(id=43438)