GPT-4V

GPT-4V is OpenAI’s multimodal extension of GPT-4, publicly detailed on September 25, 2023, that accepts text and image inputs—up to multiple images per prompt—and outputs text only. It offers a 128,000-token context window and processes images up to 20 megapixels, with the API endpoint `gpt-4-vision-preview` charging roughly $0.00765 per 512×512 low-res image tile and $0.01145 per high-res tile as of late 2023. On the MMMU validation benchmark, GPT-4V initially achieved 56.8% before later revisions, outperforming Gemini Pro Vision 1.0 (47.9%) and open-source LLaVA-1.5 (36.1%) at its launch. The model handles optical character recognition, object localization, chart analysis, and dense image description without generating images. Its immediate importance stems from being the first large-scale vision-capable model integrated directly into ChatGPT Plus and the API, enabling enterprises to automate document parsing, UI navigation, and visual Q&A in production, which accelerated the shift from text-only AI to practical multimodal deployment and intensified the race with Google’s Gemini and open-source visual-language models.

Two-hop subgraph: this entity, every entity it directly relates to, and every entity those neighbors relate to. Drag a node, scroll to zoom, click to inspect — or click any neighbor and re-center the atlas there.