Basis completes a tax workbook 2x faster with GPT-6 Astra

Basis, a company that develops AI agents to automate manual accounting tasks, has evaluated OpenAI's GPT-6 Astra by comparing it to GPT-5.6 Sol on a complex 50-tab tax workbook. According to the company, GPT-6 Astra completed the workbook twice as fast while demonstrating a stronger understanding of user intent. Basis noted that the model makes better decisions at the start of tasks, allowing agents to take more direct paths and spend less time correcting mistakes. Additionally, GPT-6 Astra can dynamically adjust the amount of reasoning it uses as a task progresses, dialing up computation for difficult steps and using less for easier ones while maintaining its cache. This capability helps reduce costs and response times, making long-running tasks more economical. Basis reported about a 20% improvement in internal evaluation scores driven by the model's better understanding of user intent, including knowing when to ask questions, flag assumptions, and follow instructions. Furthermore, GPT-6 Astra can infer expectations from a broader context with fewer explicit instructions, reducing the need for Basis to write rules for individual situations and increasing confidence that agents can handle unforeseen scenarios.