AI news · October 12, 2026
Nace opens Drex 1.5 for probability-based option scoring
Nace's Drex 1.5 returns a probability for each option in one pass and scored 58.28 on Nace's Decision Index; it is a direct alternative to asking a chat model to choose.
- forward pass for an option score
- 1
- Drex score on Decision Index 0.2.1
- 58.28
- models on that index
- 71
Nace is offering Drex 1.5 as a decision model rather than a general chat assistant. Its developer page shows an API, a command-line and MCP route, and Hugging Face weights for self-hosting. The model returns a probability for each option in one forward pass, which suits ranking, routing and structured choices better than a long answer that happens to end with a recommendation. Nace's Decision Index 0.2.1 lists Drex at 58.28, in the top five of 71 models scored on September 28.
That index is useful context, not proof that Drex will choose well for your business. A standard language model with a forced JSON output may be simpler for choices that need explanation, while Drex could be cheaper and more stable for repeated classification. Try it on a labelled set with abstain and tie cases. The missing evidence is calibration outside Nace's own index and across decisions where the options change over time.
What you can do with it
Use the sample API call or add the MCP server with uvx nace-mcp, then feed it a small labelled routing set. Compare probability calibration, not only top-choice accuracy, and keep the original inputs and chosen option for audit. Self-host the Hugging Face weights only after the hosted path proves useful.
Our take
Drex is a clean response to a common product mistake: using a chat model for a choice that should be scored. Its value depends on calibration and stable categories, and the public index is not an independent business test. Builders with routing or triage workloads should run a paid API trial before redesigning their stack.
Links Try it
Source: Nace AI — Made With Models writes the brief; the reporting is theirs.