Figma has confirmed a partnership with Anthropic to launch Code to Canvas, a feature designed to turn code generated by AI tools into fully editable design files inside Figma. The new tool is aimed at teams that build interfaces with AI agents and then need to bring those outputs into a collaborative design workflow without recreating the work from scratch.
AI-generated interfaces move straight into Figma
According to Figma, Code to Canvas lets users import interfaces created through AI coding tools directly onto the Figma canvas. Once inside the platform, teams can refine the design, compare alternatives, and align on design decisions in the same environment they already use for interface work. It is a straightforward change in workflow, but a meaningful one. Frontend output that begins as code can now be edited as design.
Anthropic rolls out Claude Sonnet 4.6 alongside the partnership
Anthropic released Claude Sonnet 4.6 in connection with the Figma announcement and said the model delivers stronger performance in computer use, coding, design, knowledge work, and data processing. The company has made Sonnet 4.6 the default model for Free and Pro users in the Claude chatbot and in the Claude Cowork productivity tool.
Anthropic is presenting Sonnet 4.6 as its most capable Sonnet model to date. The release also includes a beta 1 million token context window. At that size, the model can handle an entire codebase, long contracts, or dozens of research papers in a single request, based on the company’s description. Anthropic said the model moves close to Opus-level intelligence while staying at a lower price point.
Early testers favored Sonnet 4.6 over prior models
Developers with early access reportedly preferred Sonnet 4.6 over Sonnet 4.5 by a wide margin, mainly because of better consistency and stronger instruction following. In Claude Code testing, users chose Sonnet 4.6 over Sonnet 4.5 about 70% of the time. Users said the newer model was more likely to read context before editing code and to consolidate shared logic instead of duplicating it.
Testers also frequently selected Sonnet 4.6 over Claude Opus 4.5, Anthropic’s most advanced model released in November 2025. The company said users preferred Sonnet 4.6 to Opus 4.5 about 59% of the time, describing it as less prone to overengineering and more reliable in following instructions. Reports from testing also pointed to fewer false success claims, lower hallucination rates, and steadier completion of multi-step tasks.
Benchmark and safety results were released together
Anthropic said Sonnet 4.6 improved across benchmarks and highlighted one result from the Vending-Bench Arena business simulation. In that evaluation, the model invested heavily in capacity during the first 10 simulated months before shifting its strategy toward profitability, a move that helped it finish ahead of competitors.
The company also said it ran extensive safety evaluations and found Sonnet 4.6 to be as safe as, or safer than, other recent Claude models. Safety researchers described the model as showing strong safety behavior, with no signs of major concern around high-stakes forms of misalignment. Anthropic added that early customers saw gains in frontend coding and financial analysis, while visual outputs were described as more polished than earlier Sonnet versions, with stronger layouts, animation, and design elements that needed fewer iterations to reach production quality.