Hebbia Tests Find GPT-6 Astra Leads Financial Research Tasks
The model led the next-best alternative by 13.6 percentage points in citing required source types and met 97.9% of presentation requirements.

Hebbia’s tests of GPT-6 Astra found that the model cited required source types more often than competing models and met more presentation requirements in financial research tasks.
The model led the next-best alternative by 13.6 percentage points in citing required source types. Hebbia also found that 99.5% of its cited excerpts were relevant to the questions asked.
For presentation tasks, GPT-6 Astra met 97.9% of applicable requirements, a 14.4-percentage-point lead over competitors. The tests took place in August and September 2026.
Hebbia’s Matrix platform works with large collections of documents, including customer contracts and financial statements. In a hypothetical industrial-software acquisition scenario, Matrix reviewed about 2,000 files and produced PowerPoint slides following user guidelines.
“Citations help with traceability as well as giving users confidence that the AI found the right answer,” said Kavita Anand, a member of technical staff at Hebbia.