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

OpenAI has published a case study on Asana, describing how the collaboration software company cut the running costs of its browser agent.

In the internal browser tests cited by the post, Asana reduced model costs by a factor of 76 while making the agent 5 times faster. Both results came from the same round of testing, suggesting the gains were about efficiency as much as thrift.

The headline credits GPT-6.1 Sol with the cost reduction, while the summary says the team used GPT-6 Astra inside Codex to do the work. Either way, the point is that stronger model capability was wired into the browser agent's workflow.

According to the post, the payoff is that Asana can now offer customers more capable models without letting inference costs dictate the product experience.

For teams pushing agents into production, the case is a reminder that cost optimization is itself a product decision rather than a purely backend concern.

What to watch next is whether these test results hold up under larger volumes of real traffic, and where Asana reinvests the savings.