Improved Agents Efficiency
Published
Framer Agents now use far fewer credits on larger, multi-step builds, giving you more room to create pages, make site-wide updates, and refine your work within the same credit budget. We worked directly with OpenAI to improve model efficiency in Framer, resulting in OpenAI models using around 25% fewer credits on multi-step builds. Alongside that work, we improved all models to complete tasks in 15–30% fewer steps, use 30–40% fewer tokens, and save another 10–15% through better caching. Agents now also catch and resolve more unintended side effects.
We’ve also made GPT-5.6 Terra the default model in Framer. In our testing, it is more capable than GPT-5.5, twice as fast, and uses 37.5% fewer credits. Combined with the broader efficiency improvements, multi-step builds with Terra use 60–67% fewer credits than GPT-5.5 did when Framer Agents launched. Sonnet 5 and Opus 4.8 benefit from the same improvements, using 48% fewer credits. These improvements are now live across Framer Agents and External Agents.