Nace releases Drex for Finance decision model
Nace has made Drex for Finance generally available, a fintech-tuned decision model that picks only from options the caller supplies and outscores TypeSafe, OpenAI and Perplexity rivals on Nace's own finance index.
Nace announced Drex for Finance, a version of its Drex Max model trained further on financial decisions. The model is generally available. Callers send it a case, such as a transaction, an invoice, a support message or a filing excerpt, together with the questions to answer and the allowed options for each. The model picks one option and returns a probability for it. Because it can only answer from the options listed, Nace says it can't invent an answer.
Nace tested the model on 48 public finance test sets that it says were held out from training. Drex for Finance scored 77.3 on the company's Drex Finance Index. Jev 1.14 from TypeSafe scored 63.6, Perplexity's pplx-decider v1.2 scored 65.0 and OpenAI's GPT-6 Luna scored 58.4. Drex for Finance got 85.7% of 3,361 decisions right, and its median response time on a short case was 0.82 seconds. The index scores skill above chance and sets a model's score to 0 on any set where it does worse than a coin flip.
The model led in 10 of the 11 rows Nace reported. The largest margin over Jev 1.14 was in news and headlines, at 23.1 points. The one category where it did not lead was markets and reasoning: it scored 41.5, behind pplx-decider v1.2 at 43.1 and Drex Max v1.1 at 43.0. Nace said the finance training "didn't move it." Absolute scores stayed low in credit and risk (48.3) and monetary policy (53.8).
Nace says general performance held up. On JevBench the score rose from 93.9% to 94.4%, and on Decision Bench it rose from 94.2% to 94.6%.
The model ID is drex-max-finance, and a request can be up to 262,144 tokens long. The announcement does not give pricing, and no outside party has checked the results.