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

What happened

Enterprise application provider Basis conducted tests measuring large model performance for tax industry use cases, finding that GPT-6 Astra completes a tax workbook with 50 tabs twice as fast as the prior-generation GPT-5.6 Sol. GPT-6 Astra also demonstrated significantly improved user intent understanding, which has substantially increased Basis’s confidence in deploying the model for real business scenarios.

Key facts

Test conductor

Basis

Tested model

GPT-6 Astra

Baseline comparison model

GPT-5.6 Sol

Test task

Processing of a tax workbook containing 50 tabs

Measured performance

Processing speed is 2x that of the baseline model

Why it matters

These test results show that large models are delivering steadily improving efficiency for structured professional document processing tasks. For developers building tools for the finance and tax sector: the latest model generation promises more powerful, streamlined productivity tools. For the broader industry: the findings confirm that large model iterations are increasingly aligned with real business efficiency requirements, providing a reference benchmark for deployments in professional service domains.