Asana cuts browser-agent model costs 76x
An OpenAI case study says Asana cut estimated model costs 76x and ran browser tasks 5x faster by optimizing a GPT-6.1 Sol agent workflow, showing how efficiency work can change which models enterprises can afford to deploy.
Asana has cut the estimated model cost of its browser-agent runs by 76x, according to a customer case study published by OpenAI. The case study says the optimized workflow, built on GPT-6.1 Sol, also runs browser tasks 5x faster and costs an average of $0.47 per run in estimated model spend.
The case study lists Codex as the OpenAI product involved. It credits GPT-6 Astra, running inside Codex, with doing the experiments. An engineer set the direction, and the results went into production through a system the case study calls "Command." One quote, not attributed to a named person in the material, says the work "would have taken me one to two months by hand," but took about a week with Astra in Codex. The engineer would set a /goal before going to bed and review the results in the morning.
Another quote describes the business case: "Cost used to limit which models we could offer customers for these workloads. By making the agent more efficient, we can give customers a better, faster model while lowering our operating costs."
The material does not give the baseline model or workflow behind the 76x and 5x comparisons. It also does not say what the browser agent does, how many runs were measured, or what accuracy or quality metrics were tracked. All figures are labelled as estimates, and the source is vendor marketing, not an independent evaluation.
The case study ends with a broader claim: "Shipping speed is no longer the bottleneck; human attention is."