Drex 1.5
A 9-billion-parameter decision model open-sourced by Nace AI. Rather than generating text, it produces probability-based quantitative scores over candidate options. It supports a 128K context window and can score multiple options within a single forward pass.
Timeline (last 90 days)
Nace.AI open-sourced a decision model called Drex 1.5
Drex 1.5 is a decision model with 9 billion parameters (9B)
Drex 1.5 returns a probability value for each candidate option rather than generating natural-language text
Drex 1.5 can return probabilities for all candidate options in a single forward pass
Drex 1.5 achieved a score of 58.08 on the Decision Index 0.3.1 benchmark
Drex 1.5 supports a context reading capability of up to 128K tokens
The score of 58.08 lacks sufficient horizontal reference points, as there are no publicly available scores from a large number of other models on the same benchmark at the same version for comparison
Drex 1.5 is suitable for tasks such as risk assessment, option ranking, A/B candidate scoring, and tool or action selection in agent scenarios
When using a generative model to evaluate K candidate options, the total cost grows linearly with the number of options, whereas a single forward pass can compress the inference cost to a single computation
Being open-source allows users to customize scoring logic for their business, continue training on private data, and provides stronger controllability and auditability
All Facts (10)
Being open-source allows users to customize scoring logic for their business, continue training on private data, and provides stronger controllability and auditability
50%UnverifiedDrex 1.5 is a decision model with 9 billion parameters (9B)
50%UnverifiedDrex 1.5 returns a probability value for each candidate option rather than generating natural-language text
50%UnverifiedDrex 1.5 can return probabilities for all candidate options in a single forward pass
50%UnverifiedDrex 1.5 achieved a score of 58.08 on the Decision Index 0.3.1 benchmark
50%UnverifiedDrex 1.5 supports a context reading capability of up to 128K tokens
50%UnverifiedThe score of 58.08 lacks sufficient horizontal reference points, as there are no publicly available scores from a large number of other models on the same benchmark at the same version for comparison
50%UnverifiedDrex 1.5 is suitable for tasks such as risk assessment, option ranking, A/B candidate scoring, and tool or action selection in agent scenarios
50%UnverifiedWhen using a generative model to evaluate K candidate options, the total cost grows linearly with the number of options, whereas a single forward pass can compress the inference cost to a single computation
50%UnverifiedNace.AI open-sourced a decision model called Drex 1.5
50%