hybrid
taxonomy axis: compute_structure
Decision rule: Neither pattern reaches 80% of layers (e.g. a substantial mix of dense and MoE layers).
Anchor: None in atlas yet — first qualifying report sets the anchor
Models in this category (3)
| Model | Org | Date | Block | Attention | Context | Total | Active |
|---|---|---|---|---|---|---|---|
| Jamba | AI21 Labs | 2024-03-28 | hybrid | GQA | 1M | 52B | 12B |
| Qwen3-Coder-Next | Qwen Team | 2026-02-28 | hybrid | hybrid | 262,144 | 80B | 3B |
| Qwen3.5 | Qwen Team | 2026-04-21 | hybrid | hybrid | 262,144 | not disclosed | not disclosed |
What varies within this category
attention variant varies — GQA, hybrid. trained context varies — 1000000, 262144. openness varies — open-weights, undisclosed. scale class varies — frontier, large.