β Catalog
Arete-27B
AvailableNous ResearchOpen source
Supervised fine-tune of Alibaba Qwen3.8-27B by Nous Research, trained only with SFT (no RL or architecture changes) on about 52.8 million tokens at 64k context. Vision encoder left frozen. Aimed at changing political framing relative to the base model; Nous reports propaganda answers falling from 22% to 0.9% on its 803-prompt eval and Hermes Bench moving from 65.77% to 65.60%. Apache-2.0 weights, training data and eval are public.
Specifications
- License
- Open source Β· Apache-2.0
- Weights
- Downloadable
- Architecture
- Dense
- Parameters
- 27B
- Context window
- β tokens
- Max output
- β
- Knowledge cutoff
- β
- Price (in / out, $/M)
- β
- Modalities
- Text
Benchmarks
No benchmark scores recorded yet. Spotted some? Submit a correction.
Vendor-reported figures are claims until independently verified. See methodology.