Expired60% confidenceFactExact time
Current multimodal large models including GPT-4o and Claude 3.5 Sonnet can 'see' images but encode them into visual tokens processed at a semantic level rather than perceiving pixel-level details
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026(expired)
Sources
Cursor Fails at UI Design Reproduction: The Real Capability Boundaries of AI Coding
bilibili明天玩点什么6/3/2026
Related Claims
Unverified在多模态大模型（如GPT-4o、Claude 3.5 Sonnet）中，图片输入会被转换为token进行计费，一张标准网页截图可能消耗数百到数千个token79% similarVerifiedThe visual understanding of multimodal large models such as GPT-4V, Gemini Vision, and Claude is built on an architecture that deeply integrates a visual encoder (typically a Vision Transformer) with a large-scale language model.76% similarVerifiedGPT-4 Vision Preview added a vision encoder module on top of the purely text-based GPT-4, requiring images to first be processed by the vision module before being passed to the language model.76% similarUnverifiedLLM Visualization (bbycroft.net) uses 3D animation to display the complete data flow of GPT-class models, from token embedding to final output, with every step viewable interactively.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/53566API
curl https://kongchang.com/api/v1/knowledge/claims/53566MCP
get_claim(id=53566)