OpenAI, Thrive Holdings launch 97% accurate self-improving tax AI

Editorial illustration for: OpenAI and Thrive Holdings develop self-improving tax AI with 97% accuracy

In brief

  • OpenAI and Thrive Holdings developed AI system drafting tax returns with 97% accuracy.
  • Pilot program processed 7,000 returns across 30+ accounting firms.
  • Practitioners saved one-third preparation time and achieved 50% throughput gains.
  • System learns from accountant corrections and improves iteratively.

Pilot Results Show Significant Efficiency Gains

A pilot program ran through Crete Professional Alliance, a network of over 30 accounting firms, and processed 7,000 tax returns focused primarily on 1040 and 1041 filings. Practitioners using the system reported saving roughly one-third of their preparation time, while throughput increased by approximately 50%.

The gains accelerated as the system learned. At launch, only 25% of returns hit 75% correct field completion. Six weeks later, that number had climbed to 86%. This improvement curve reflects the system's core design: every time a human accountant fixes something the AI got wrong, the system learns from that correction and updates itself.

How the System Works

The Tax AI is built on OpenAI's Codex technology. It targeted medium- to high-complexity returns involving messy documents and K-1 data, the kind of filings that typically demand the most manual effort from tax professionals.

The feedback loop is what sets this apart. Rather than shipping a static model, the system continuously ingests corrections from human accountants and refines its field completion and accuracy. This approach means the AI doesn't just help practitioners today—it gets measurably better at helping them tomorrow.

Ownership and Next Steps

Thrive Holdings owns the resulting intellectual property and products from the collaboration. The partnership signals OpenAI's move beyond pure language models into vertical-specific applications where domain feedback can drive rapid improvement. For accounting firms managing thousands of returns per season, tools that cut prep time by a third and boost throughput by half represent material operational gains.