Grok Image 2

Grok Image 2 pairs a strong vision encoder with a careful language backbone so it can look, read, and reason in one pass. You can pass photos, scans, tables, dashboards, or UI screenshots alongside a prompt, and it extracts small text, preserves page structure, and ties explanations to the right regions for traceable answers. Multi-image threads stay coherent across pages or states, responses can be emitted as schema-true JSON for automation, and native function calling lets agents crop regions, fetch metadata, or query retrieval during a reply. The model is tuned for low latency and long contexts, which makes it practical for document automation, chart and dashboard analysis, screenshot and UI understanding, multimodal RAG, and developer copilots that need grounded visual reasoning at production speed.