A major cost drop in Asana's browser agent
Asana ran its browser-based AI agent on GPT-6 Astra inside Codex, cutting model costs by 76x in tests. In the same tests, the agent also ran 5x faster.
The company says it will use this efficiency to offer customers more capable models. In other words, the savings are not a price cut but a way to deliver a stronger agent experience on the same budget.
Why it matters
Model inference cost is one of the biggest expenses in agent-based AI products. Asana's result shows that the right model and infrastructure choices can bring that cost down dramatically.
Case studies like this offer a concrete answer to whether enterprise software companies can ship AI agents at scale.
