geoagencies

Hebbia tests GPT-6 Astra on financial research citations

Hebbia has published results from tests of a prerelease build of OpenAI's GPT-6 Astra, run between August and September 2026 on the same financial research cases and grading criteria for each model. Domain experts wrote or audited the criteria against source documents.

In those tests, GPT-6 Astra scored 13.6 percentage points above the best alternative on citing the required types of sources, and 99.5% of the excerpts it cited were relevant to the question. On presentation work it met 97.9% of applicable requirements, 14.4 percentage points ahead of the best alternative, and ranked first on both following the brief and slide quality.

Hebbia's Matrix links answers back to the documents behind them, and users can open a citation to see the underlying passage, with slide footnotes leading back to source material. Kavita said traceability matters because users base millions of dollars of decisions on what the model returns, and that even a single percentage-point gain can yield marginal benefit across a data room of thousands of documents.

For brands, the results point at the same thing that matters in AI answers generally: being the source a model selects and cites, not just being mentioned. Hebbia's focus is on which documents and cells support an answer, which puts the emphasis on the traceable source rather than the generated text.

Reported by OpenAI News ↗

Topics: Citations, ChatGPT

Related