LONDON, UK / Algosoup today announced algofin 1.0, an AI model trained for financial decisions and licensed for financial firms to run on their own infrastructure. On published benchmarks, it beats leading classifier models by 11 points on financial sentiment and matches it on banking intent and scam detection.
The model reads a question once, such as a customer message, a news story or a report, and gives a probability for each possible answer. Firms can deploy algofin 1.0 within their own networks, including in air-gapped environments, on a single GPU.
Financial decision performance
Algosoup scored algofin 1.0 with the benchmark team’s published evaluation code, unmodified, after first reproducing their published numbers for the base model. All three tasks are scored zero-shot: the model was not trained on any of the benchmark’s data.
“Finance firms should be able to own the model that makes their decisions, and decide where it runs.” said Alex Reid, Co-founder and CEO of Algosoup.
Deployment under the firm’s control
algofin 1.0 supports self-hosted and air-gapped deployment. Financial firms can operate the model inside their network, including in the UK, and choose when to introduce model updates.
Track Investments“Public benchmarks will get you into the room. What really matters is how the model handles a firm’s own messages, filings and edge cases, and because it runs on their hardware, they can test that without their data leaving the building,” said Henry Qin, Founding Partner at Algosoup.
Financial firms interested in licensing algofin 1.0 can contact Algosoup to discuss evaluation and deployment requirements.
About Algosoup
Algosoup is a London-based forward deployed engineering company building production systems across artificial intelligence, machine learning, computer vision and data. Its forward-deployed engineers embed with client teams to design, build, deploy and operate software. Algosoup works with organisations across the United Kingdom and Europe, including in financial services and other industries.
