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

Asana used GPT-6 Astra in Codex to optimize StackAI's browser agent by testing caching, history retention, and screenshot policies across frontier models. The optimized workflow on GPT-6.1 Sol averaged $0.47 and about four minutes per run, 76x cheaper and 5x faster than the original production setup. For Model B, optimization cut estimated cost from at least $36.21 to $1.24 per run, a 29x reduction. Caching 89% of input at 5% of the uncached price made each call about 3x cheaper, and a larger history budget improved answer completion from 3 of 18 runs to all 18. Asana has released the browser navigation changes in StackAI and plans to incorporate similar testing into platform evaluations.

(Source：OpenAI)