Your invoices don't need to leave the building

A demo from open-source project NobodyWho found that the local Qwen 3.5 0.8B model extracted structured information from 5 invoices about 3x faster on average than Anthropic's Claude Haiku, at a total cost of 0 EUR and with data kept 100% private. The comparison used the FATURA2-invoices dataset and ran through NobodyWho, with the demo code published on GitHub.

← Back to blog https://www.nobodywho.ai/posts/ Your invoices don't need to leave the building For a non-technical audience, it's hard to grasp the value of open models. Or should we say private models? The best way of demonstrating value is to start with a very practical use case. That's what we presented on one of our latest demos. Extracting structured information from invoices. That's a problem a lot of companies have - right? For this quick demo we went with a super straightforward task. 5 invoices https://huggingface.co/datasets/mathieu1256/FATURA2-invoices , and a single goal: Extract structured information from them. We put Qwen 3.5 0.8B https://huggingface.co/NobodyWho/Qwen Qwen3.5-0.8B-GGUF vs. Anthropic's Haiku https://www.anthropic.com/claude/haiku model. The results are surprising. Not only is the local model 3x faster on average. It also costs a total of 0 EUR. All while keeping your data 100% private. And everything is powered by NobodyWho https://github.com/nobodywho-ooo/nobodywho . If you want to run the demo yourself, it's just a couple of commands, and the code is right here https://github.com/nobodywho-ooo/nobodywho-invoice-demo . Everything NobodyWho does is open-source, please leave a star on GitHub https://github.com/nobodywho-ooo/nobodywho to support us ❤️ Get new posts in your feed reader: subscribe via RSS https://www.nobodywho.ai/feed.xml .