Framer Cuts AI Agent Costs by a Quarter and Adds Accessibility Controls

Framer made GPT-5.6 Terra the default model for its Agents feature on 21 July 2026, working directly with OpenAI on the switch, and says it cuts credit use by roughly 25 percent on multi-step builds compared with the model it replaces. Against older versions of Framer's Agents, the company puts the saving as high as 60 to 67 percent.

For studios using Framer's Agents to scaffold layouts or wire up interactions from a prompt, that is a direct cost reduction on work that was already billed by credit consumption, not a vague efficiency claim. A cheaper default model on multi-step builds means more iteration is affordable within the same budget, which matters most on exploratory work where the first few attempts are rarely the ones that ship.

Framer followed the model change with a feature update on 28 and 29 July that is more about capability than cost. Agents can now edit Page Effects directly, set accessibility layer attributes without a manual pass afterwards, and adjust Text Shadow properties. The update also added CMS video insertion and improved handling of Figma imports, specifically around Overflow: Clip behaviour, which has been a persistent source of layout drift when moving designs from Figma into Framer.

The accessibility attribute support is the change worth paying attention to beyond the headline cost cut. Letting an agent set A11Y attributes as part of the build step, rather than treating accessibility as a separate audit pass at the end of a project, is the kind of default that actually changes outcomes, because it removes the step most commonly skipped under deadline pressure.

None of this is a dramatic reinvention of what Framer does. It is a tool getting cheaper to use and slightly more thorough at the same time, which is a fairly rare combination, and one worth noticing given how many AI feature updates land as pure cost increases dressed up as capability.