Basis, an AI agent company for accountants, evaluated OpenAI's GPT-6 Astra against GPT-5.6 Sol on a complex 50-tab tax workbook task. The test showed that GPT-6 Astra completed the work in half the time and demonstrated a deeper understanding of user intent.
- GPT-6 Astra took 50% less time to finish the 50-tab workbook compared to GPT-5.6 Sol.
- Basis observed approximately a 20% improvement in internal evaluation scores, driven by better intent recognition.
- The model dynamically adjusts reasoning computation based on task difficulty while maintaining its cache.
- GPT-6 Astra can infer expectations from broader context, reducing the need for explicit rules.
The dynamic adjustment of reasoning helps reduce cost and response time, making long-running tasks more economical for Basis and its customers.