I’m open sourcing an autonomous financial research and analysis agent. The agent can search SEC filings, extract key financials, and build financial models.
It’s scored 80% on the public Finance Agent validation set with GPT-5, compared to the top result of 55% listed on their website for their private validation set.
But, their public validation set has mistakes. There are quite a few cases where the “ground truth” answers in the benchmark are wrong. I’ve documented each case with citations directly to SEC Edgar here (https://github.com/lucasastorian/intellifin-agent/blob/main/...)
Accuracy with GPT-5 jumps to 92% once we fix those mistakes in the eval.
You can clone the repo and rerun the benchmark yourself.
Next step should be turning this into an open-source “Cursor for Finance,” with a proper UI for equities research and financial modeling.
I’m open sourcing an autonomous financial research and analysis agent. The agent can search SEC filings, extract key financials, and build financial models.
It’s scored 80% on the public Finance Agent validation set with GPT-5, compared to the top result of 55% listed on their website for their private validation set.
But, their public validation set has mistakes. There are quite a few cases where the “ground truth” answers in the benchmark are wrong. I’ve documented each case with citations directly to SEC Edgar here (https://github.com/lucasastorian/intellifin-agent/blob/main/...)
Accuracy with GPT-5 jumps to 92% once we fix those mistakes in the eval.
You can clone the repo and rerun the benchmark yourself.
Next step should be turning this into an open-source “Cursor for Finance,” with a proper UI for equities research and financial modeling.
Feedback & questions are welcome.
https://github.com/lucasastorian/intellifin-agent