Drex Max v1
Nace AI's open-weight 27B dense decision model that returns probabilities over up to 255 options instead of free text.
This entry is marked for review. Check the original sources before relying on its details.
Overview
Drex Max v1 is a dense 27B-parameter decision model from Nace AI, open-sourced around November 1, 2026 under the Open RAIL-M license. Instead of generating text it scores up to 255 answer options per question, with a 262,144-token context window. Nace published bf16 weights, a Q8_0 GGUF, an affine Q4 checkpoint and forks of llama.cpp and Ollama; a 4-bit build is reported to run on 24 GB Apple Silicon Macs. Benchmark figures (such as a Decision Index of 67.52) are vendor-reported and not independently reproduced, and one source disputes that weights are downloadable.
This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.
Key features
- Dense 27B parameters, 262,144-token context
- Scores up to 255 options per question rather than generating text
- bf16, Q8_0 GGUF and Q4 checkpoints plus llama.cpp and Ollama forks
- 4-bit build reported to run on 24 GB Apple Silicon
- Open RAIL-M license (use restrictions apply)
Use cases
- Classification and routing decisions with calibrated option probabilities
- Local on-device decision scoring on Apple Silicon
- Choosing among candidate actions in agent pipelines
Pricing
Free open weights under Open RAIL-M; hosted access also listed on OpenRouter (pricing unknown).
Pricing reflects the source record and may change. Confirm current terms with the provider.
Dates and sources
- Released
- 1 Nov 2026
- Record updated
- 6 Nov 2026
- Record last checked
- 6 Nov 2026