Asana Cuts Model Costs with GPT-6 in Browser Tests
◆ The brief
Asana, a major workplace productivity software company, has successfully cut model costs and boosted performance by testing GPT-6 Astra within its coding environment. By integrating this advanced artificial intelligence model into a system called Codex, Asana made its browser-based agent significantly cheaper and faster. An AI agent is a software program capable of executing complex, multi-step workflows on its own, while browser agents specifically navigate web pages to complete tasks for users. In internal tests, the updated system achieved a dramatic speed improvement and a massive drop in operating expenses. This breakthrough demonstrates how rapidly foundational AI models are becoming more efficient, allowing companies to offer powerful capabilities without breaking the bank. Efficient models make it commercially viable to deploy automated assistants to millions of everyday workers for routine office tasks. Moving forward, software companies are expected to rapidly integrate these cheaper, faster models into their platforms to stay competitive in the enterprise market.
◆ Key points
- 01Asana tested GPT-6 Astra inside its Codex environment
- 02The browser agent became 76 times cheaper to run
- 03Performance tests showed a 5x increase in speed
- 04Greater efficiency makes enterprise AI much more practical
◆ By the numbers
Making artificial intelligence faster and dramatically cheaper removes the biggest financial barriers to widespread adoption. This efficiency allows companies to embed powerful automation into everyday workplace software, changing how millions of people work.