Asana cuts model costs 76x in browser tests with GPT-6.1 Sol

The improvements stem from three changes introduced by StackAI CTO Frank Hidalgo: extending caching to include browsing history, enlarging the text‑retention budget, and removing screenshots in batches instead of at every step. In a 144‑run study across four frontier models, GPT‑6.1 Sol completed all 18 test runs with correct answers, while other models often failed or hit step limits. The new policy also reduced cost from $1.97 to $0.47 per run, with 89 % of input coming from cache at just 5 % of the uncached price. Asana has shipped these changes to customers and plans to embed the testing framework into StackAI for ongoing cost‑runtime‑quality comparisons.