Anthropic has updated Claude to generate interactive web pages inside canvas graphics, a capability that mimics the old Flash era while retaining real HTML functionality. Felix Rieseberg, an engineer at Anthropic, confirmed the feature on X, noting it allows the model to embed clickable buttons and selectable text within complex visual effects like knitting patterns or 3D cubes. Previously, developers faced a trade-off where standard HTML elements were accessible but visually rigid, while canvas offered visual freedom at the cost of interactivity and screen reader support. This update places actual page elements inside the canvas layer, solving that conflict. The system now renders content as real text that users can highlight and copy, even when surrounded by dynamic animations or warped backgrounds.
The change matters because it expands what large language models can produce without relying on static screenshots or external image generators. Google is simultaneously developing HTML-in-Canvas as a web standard to ensure this hybrid approach works across different browsers. Support is currently limited to Chrome and Edge from version 157, as well as specific desktop versions of Claude. Safari and Firefox do not support the feature yet. Users must explicitly request HTML-in-Canvas in their prompts to trigger the new rendering mode. This tool does not alter how Claude handles privacy, API limits, or customer account data, but it does increase the visual complexity of outputs generated directly in the chat interface.
- Support is available in Chrome and Edge starting with version 157
- Safari and Firefox currently lack compatibility for the feature
- Users must explicitly request HTML-in-Canvas in their prompts



