GPT-5.6 is connected to Figma Make: the last mile from design to code is accelerated

On July 9, 2026, Figma announced that GPT-5.6 for OpenAI is now available in Figma Make. The official reason given is very straightforward: GPT-5.6 has accurate first-time generation results and rapid iteration capabilities, which can provide a strong starting point for development and construction in Make, making the process "from design to code" smoother.

A typical ecological horizontal integration

This is a typical "design platform × cutting-edge large model" capability splicing. For designers and developers, the value is that they don’t have to leave the Figma canvas context to switch to external models—the design draft, component semantics, and code generation are completed in the same workflow. The positioning of GPT-5.6 in Make is not a "general chat", but a "start for development and construction", which emphasizes the starting point of generation and the speed of iteration.

Putting it into the evolution sequence of Figma, the signal is clearer: Figma has previously launched Figma MCP Server (for external agents to access design assets) and Figma Make (to connect the code base with the canvas), and now it is connected to GPT-5.6 - the three pieces of the puzzle put together, Figma is evolving from a "design tool" to an "AI infrastructure between design and development".

Model open or model neutral?

From an industry perspective, what is noteworthy about this cooperation is the "right to choose models." Figma Make's access to GPT-5.6 is a starting point. Whether more models such as Claude and Gemini will be opened in the future will determine whether it is a "neutral AI infrastructure" or a "channel for a certain model." Referring to the multi-model strategy of the AI ​​programming track, the model openness of design tools will most likely become the focus of competition.

For domestic design collaboration tools, Figma's "MCP + Make + Large Model" three-layer architecture is a replicable template: first use MCP to allow external agents to read design assets, then use generated links to open up the code, and finally use multi-model selection to retain users. The gap between design and development is being filled step by step by this "infrastructure" approach.

Several directions worth tracking in the future:

1. Make’s rhythm of supporting more models: when will Claude and Gemini be connected, and how open they will be.
2. Actual measurement of first generation quality: Performance of GPT-5.6 in real design draft to code scenarios.
3. MCP Server adoption: Whether external agents (such as Cursor, Copilot) will become callers of Figma assets.
4. Design to code boundary: Whether the development scenarios covered by Make can move from prototypes to production-level front-ends.