#Drex

Small verification models auditing frontier-model or OCR output point to a broader pattern: routing doubts through a cheap checker beats paying for a large model on every token. Teams processing invoices, forms, or scanned archives at volume can adopt the same verify-and-retry loop instead of scaling up model size. Compared with running GPT-6 Astra (low) at US$70.57 per 1,000 pages, the tenfold cost gap makes this relevant to startups and enterprise back offices across ASEAN where document digitisation budgets are tight.

rohanpaul_ai@rohanpaul_aiSource

Calibration training—tuning confidence to match real accuracy—is the detail that separates this from typical fine-tuning writeups, since miscalibrated probabilities are a common failure point for models used in decision support. Practitioners evaluating open-weight models for document-heavy workflows can compare Drex's approach against closed APIs and test it directly in Nace Console. The full training methodology is public, making it a useful reference for teams building constrained-answer systems.

TestingCatalog@testingcatalogSource

The interesting part is the interface shift: instead of generating text, these models emit calibrated probabilities over fixed options, which suits scoring, routing and agentic decision points where latency matters. Builders assembling agent pipelines or classification backends can try the weights on Hugging Face or read the paper at arxiv.org/abs/2611.07854. Compared with calling Claude Opus 5.5 or GPT-6 Sol, Drex trades text generation for sub-200ms structured decisions under an open licence.

elvis@omarsar0Source

The 128k context window eliminates preprocessing for long documents, a significant advantage over previous decision models. Developers can now handle entire legal or video analysis tasks in a single pass, with a cost-effective option at $0.04 per 1M tokens. This model's top ranking on the Decision Index makes it a compelling alternative for businesses needing automated decision systems.

Ate-a-Pi@svpinoSource

The implication is that generation is wasted compute when the required output is a choice among known options, and a small classifier-style model can serve that role at a fraction of the cost. Teams building agent routing or tool-selection layers can swap a generative model for Drex in those steps while keeping a larger model for prose tasks. Compared with typical 100B-plus LLMs used for routing, the sub-6B size points to lower latency and cost.

rohanpaul_ai@rohanpaul_aiSource