OpenAI published on October 9, 2026 that Asana integrated GPT-6 Astra (referred to as Codex 6.1 Sol) into their browser agent, with impressive results. In the tests described on OpenAI’s announcement, Asana reduced large language model operation costs by a factor of 76 and improved response latency by 5 times, compared to prior solutions.
Key practical changes:
- The agent still runs in the user’s browser, but leverages GPT-6 Astra for NLP and understanding user requests.
- Costs have dropped so significantly that more advanced models can run per user at scale, enabling richer customer features.
- Performance boosts mean more real-time context handling in the browser, without sacrificing speed.
If you’re developing browser-based agents, these findings suggest that integrating newer, cost-efficient LLMs like Astra can enable deployment of powerful models to end customers while controlling infrastructure costs. You should review your agent’s current model selection and consider experimenting with recent OpenAI releases for both cost and speed gains.