Compare AI Models: Pricing, Context & Benchmarks | OpenRouter

Open original

Drex Max Finance is a 27B multimodal model from Nace.AI, tuned for banking, accounting, payments, insurance, and compliance work. It takes text, JSON, or document images like invoices and statements, then returns calibrated probabilities for each answer option in one forward pass, with no text generation.

  • It outputs a calibrated probability for every option on yes/no, multiple choice, or ordinal questions, in a single forward pass rather than generating text.
  • It accepts documents as input (invoices, receipts, statements) alongside structured data and plain text.
  • It supports contexts up to 262K tokens.

Drex Max Finance is a finance-specialized build of Nace.AI's Drex Max, which is described as a 27B-parameter multimodal decision model. The core idea is that you hand it a state, which can be plain text, JSON, or images like invoices, receipts, and bank statements, along with a typed question. Those question types are yes/no, multiple choice, or an ordinal score. In return, it gives a calibrated probability for each possible answer in a single forward pass. The key detail is that it does not generate text to reach that answer. That makes it a classifier-style scorer rather than a chatbot, which matters if you want consistent, machine-readable outputs instead of prose you have to parse. The target use cases are banking, accounting, payments, insurance, credit, compliance, regulatory filings, and markets. The context window reaches up to 262K tokens, which is large enough to feed in long filings, multi-page statements, or extended transaction histories in one go. Because the output is a probability distribution over defined options rather than free text, it fits pipelines where you need to threshold decisions, route cases for review, or track calibration over time. The article doesn't go into benchmarks, pricing, or how the calibration was validated, so anyone evaluating it for a production workflow would need to test those things directly. What the piece leaves out is as important as what it includes. There are no published accuracy figures, no comparison against existing document-understanding or fraud-scoring systems, and no detail on training data or licensing terms for financial data. Since the description is largely promotional, treat the claims of calibration and single-pass scoring as things to verify on your own data. If you're bookmarking this to revisit, the useful question to answer later is whether the probabilities hold up on held-out examples from your own domain.

Ask about this article