# GitHub Copilot Vision Copilot Vision became generally available on 1 July 2026. You can attach images and PDFs to a [[GitHub Copilot]] chat prompt, and Copilot reasons about them alongside your code. Available on every plan including Free, enabled by default, no admin action needed. Small feature, quietly useful, and it closes a gap that cost people a surprising number of steps per day. ## What it supports - **Files**: JPEG, PNG, GIF, WebP, and PDF - **Surfaces**: Copilot Chat in [[Visual Studio Code (VSCode)|VS Code]], on github.com, and in the GitHub Copilot CLI (attach an image path in the terminal) - **Modes**: works across ask, plan, and agent modes - **Plans**: Free, Pro, Pro+, Business, Enterprise ## Why it matters more than it looks The value is not "AI can see pictures". It is the removal of a translation step you were doing manually and constantly. A screenshot of a failing UI, a design mockup, a chart, an architecture diagram, a table in a PDF, an error dialog from a tool with no copyable text. Every one of those used to require you to describe it in prose or run it through OCR before the model could help. Now the artefact goes straight into the conversation. The agent-mode part is the interesting half. An agent that can take its own screenshots and read them closes a feedback loop that was previously open: build the UI, look at the UI, fix the UI. Without vision, the agent is writing frontend code with its eyes closed and asking you what it looks like. **The counter-argument, which is fair.** Multimodality in a coding model is often a distraction, and a specialist model would do better at images while a coding-only model would do better at code. That holds for generation. It does not hold for *input*, because the cost of switching tools to describe an image is paid on every single interaction. Convenience at the input boundary is worth a lot even when the underlying capability is unremarkable. ## Caveats - For Business and Enterprise, attachments are retained about 24 hours to provide the service. Worth knowing before dropping a screenshot with customer data in it - Quality depends entirely on the model behind the chat, and the announcement makes no claims about it - Nothing here is novel. [[Claude Code]] and [[OpenAI Codex]] users have been pasting screenshots for a while. This is Copilot reaching parity, not setting a direction ## References - Changelog — https://github.blog/changelog/2026-07-01-copilot-vision-is-generally-available/ ## Related - [[GitHub Copilot]] - [[GitHub]] - [[Visual Studio Code (VSCode)|VS Code]] - [[AI Agents]] - [[AI Tool Use]] - [[How Coding Agents Work]] - [[Claude Code]] - [[OpenAI Codex]] - [[Agentic Engineering]]