Asana Reduces Costs in Browser Testing with GPT-6.1 Sol
Updating the browser agent workflow on the StackAI platform, the company significantly lowered costs while increasing speed fivefold.
Asana optimized the browser agent within its StackAI platform using the GPT-6.1 Sol model, achieving a 76-fold reduction in model costs and a fivefold increase in processing speed.
StackAI and the Browser Agent
Through the StackAI platform it acquired, Asana helps customers build workflows that navigate websites without writing code.
Optimization Efforts
StackAI CTO Frank Hidalgo conducted comprehensive testing and research using GPT-6 Astra in Codex to make the agent faster and more cost-effective.
Codebase Analysis
GPT-6 Astra discovered that the agent cached fixed instructions but did not cache page texts or screenshot history.
Test Results and Costs
In a benchmark consisting of 144 runs, the optimized workflow on GPT-6.1 Sol delivered an average cost of $0.47 and a duration of four minutes.
Implemented Improvements
As part of the enhancements, caching for browsing history was expanded, the text budget was increased, and screenshots were managed in batches.
Process and New Features
This research process, which normally would have been expected to take one to two months, was completed in about a week, and the changes were deployed live on the StackAI platform.