- Context (input / output)
- 1.01M / 131K
- Parameters
- 2400B / 95B active
- Release date
- 2026-08
- License
- Qwen3.8-Max License
- Input modalities
- Text
- Output modalities
- Text
- Positioning
- Open-weight text foundation for Qwen3.8-Max
- Architecture
- MoE · Gated DeltaNet / Gated Attention
- Modes
- Thinking only (supports low / medium / xhigh)
- Total parameters
- 2.4T
- Exact release date
- 2026-08-12
- Experts
- 512 (10 routed + 1 shared expert per token)
- Weight status
- Open weights released (BF16)
- Default reasoning effort
- xhigh (supports low / medium / xhigh)