Nace.AI Launches Drex, a Sub-6B Decision-Only Model Topping Its Decision Index

Nace.AI (@nischaytwt) GA'd Drex, a decision-only small model, on September 25, with the news spreading via @rohanpaulai and others the next day. Drex is under 6B parameters and skips the "generate text then parse" pipeline: given facts, it outputs a probability for each candidate action in a single forward pass. Per @nischaytwt, its architecture is a small diffusion model plus RLAF, it targets Jev, and pricing is $0.04 per million input tokens.

Confirmed

• On Nace's own Decision Index 0.2 (40 benchmarks covering contracts, causality, tool calling, etc.), Drex ranks first with 51.73, a narrow 0.06-point lead that @Hesamation summarized as "beating Jev".
• It wins 23 of the 40 benchmarks; @ordax and @rohanpaulai confirm the single-forward-pass, no-parsing design.
• @Hesamation notes 250 million free tokens ship with the release.

Unconfirmed

• The official Decision Index score is marked "pending" (per @rohanpaulai and @nischaytwt).
• All results come from Nace's own leaderboard with no independent third-party validation yet.

Why it matters

• Drex carves "decision-making" out of the general LLM text-generation pipeline, trading tiny parameter count (<6B) for lower latency and cost; if independent evaluations reproduce the results, it could shape model selection for structured-choice scenarios like agent tool calling and contract review.

2026-09-25 ~ 2026-09-26 · 5 related posts

• [source] Nace Launches Drex, a Diffusion Decision Model Beating Jev on 23 of 40 Benchmarks at $0.04/M Tokens — nischay_twt · 2026-09-25
• NaceAI launches Drex, a sub-6B decision model that tops the public Decision Index at 51.73 — ordax · 2026-09-25
• [source] NaceAI's tiny Drex tops Decision Index, edges out Jev and offers 250M free tokens — Hesamation · 2026-09-25
• Nace launches Drex, a sub-6B decision model outputting option probabilities instead of prose — rohanpaul_ai · 2026-09-26

1 near-duplicate retellings: rohanpaul_ai