Basis achieves faster tax workbook completion with gpt-6 astra

Basis, a company specializing in AI agents for accounting automation, has reported that GPT-6 Astra completes complex tax workbooks 50% faster than GPT-5.6 Sol. The AI model also demonstrated a deeper understanding of user intent, leading to a 20% improvement in internal evaluation scores. Basis tested the models on a 50-tab tax workbook, noting that GPT-6 Astra made better decisions at the start of tasks, reducing the need for corrections. Troyanovsky, a representative from Basis, highlighted that the model adjusts its reasoning based on task difficulty, optimizing computation and reducing costs. This adaptability allows GPT-6 Astra to maintain its cache while improving efficiency. The model’s ability to infer expectations from broader contexts also reduces the need for explicit instructions, giving Basis more confidence in its agents’ ability to handle complex scenarios beyond internal testing. Basis evaluates its agents based on their ability to follow templates, consult tax sources, and self-check work. GPT-6 Astra’s enhanced understanding of user intent, including when to ask questions and flag assumptions, contributes to its performance. These improvements position GPT-6 Astra as a more efficient and reliable tool for automating accounting tasks, offering cost savings and faster processing for Basis and its clients.