Arete-27B vs Ling 3.2 Flash
Compare documented capabilities and access. A dash means a comparable figure is unavailable.
| Specification | ||
|---|---|---|
| Task | Language generation and alignment | — |
| Published capabilities | Fine-tuned from Qwen3.8-27B using about 52.8 million tokens for research on value alignment. | Native multimodal MoE model with 560B total and 25B active parameters for complex visual understanding, multimodal agents, 3D interaction, and medical imaging |
| Model size | 27B parameters | — |
| Access | Public weights, evaluation and training dataset | — |
| Sources | Publisher documentation | — |
| Inputs | Text | Text, Images, Video |
| Context window | — | 1,000,000 tokens |
| Input price per million tokens | — | — |
| Output price per million tokens | — | — |
| Weights | Open weights | Proprietary |
| Released | Nov 6, 2026 | Nov 5, 2026 |