Financial services needed a model that understood the weight of its words. So we built one.
Most language models learned to think by reading the internet. FinLLM learned by reading the documents that hold financial lives together. Advice records. Compliance reviews. Client conversations. The backbone of regulated work.
Over the past twelve months, Aveni worked with Lloyds Banking Group and Nationwide to turn that focus into a working model. A system trained on real documents, shaped by real interactions, and built to meet the expectations set by UK regulation.
This report captures what this year produced. It shows how the model interprets financial language with precision, how it handles the structured tasks advisers and reviewers face each day, and how a smaller, focused model managed to surpass larger general alternatives. It also includes the details that matter most: stronger recall in vulnerability detection, more accurate extraction for FactFinds.
Inside the report, you will see:
- âś“How the model was trained on curated UK financial sources, not general web text
- âś“The results from AveniBench and why they matter for real regulatory tasks
- âś“Evidence of FinLLM outperforming larger general models on accuracy, recall and structured extraction
- âś“How governance, safety and transparency were built into every stage of development
- âś“The early production outcomes that show how FinLLM behaves when the stakes are real
Inside, you will find the moments that defined this first chapter. The early experiments. The decisions that made the model more dependable in the places where dependability is not optional. Each section draws from real evidence in production, from tests grounded in actual workflows and from the quiet discipline required to build intelligence for regulated environments.
The past year laid the groundwork; the years ahead will define the legacy. FinLLM moves forward with a simple truth at its core: trust is built through careful work, steady improvement and decisions made with the people who depend on the system in mind. That is where the future of financial intelligence takes shape.
Discover how FinLLM was developed over twelve months with evidence from real production use, tested against the tasks that matter, and shaped by the people who rely on it. Read the story of FinLLMÂ
