Coding-focused offshoot of the K2 series — ~30% fewer thinking tokens than K2.6. Competitive with GPT-5.5 and Claude Opus 4.8 on coding benchmarks. 256K context, MLA attention, 1T-parameter MoE.
At a glance
Context
262K tokens
Parameters
1T MoE (32B active, 384 Experten)
Input
text, image, code
Specifications
- Status
- Released
- Date
- June 12, 2026
- Lab
- Moonshot AI
- Origin
- CN
- Type
- reasoning, code
- Open weights
- Yes
- License
- Modified MIT
- Context
- 262K tokens
- Parameters
- 1T MoE (32B active, 384 Experten)
- Input
- text, image, code
- Output
- text, code