Moonshot AI, released 2026-01-27. Sparse MoE decoder, 1T total, 32B active, 256K context. Attention: MLA.
Sparse MoEMoonshot AI 🇨🇳 · July 10, 2025
Kimi K2 (1T)
Trillion-parameter Moonshot model that essentially scales the DeepSeek V3 recipe upward.
Key detail
More experts and fewer MLA heads than DeepSeek V3.Total params
1T
Active / token
32B
Context
128K
Fact sheet
- Scale
- 1T total, 32B active (3.2% active)
- Context window
- 128,000 tokens
- Decoder type
- Sparse MoE
- Attention
- MLA
- Layer mix
- 61 MLA
- KV cache / token (bf16)
- 68.6 KiB · Low
- License
- Modified MIT License
- Released
Efficiency
Parameters active per token
3.2%
32B of 1T weights run for each token.
AA Intelligence Index v4.3.2
Source12.7
- Scientific
- 3.7
Architecture concepts
Sources & resources
Use the ← and → keys to step through releases.
Similar architectures
Moonshot AI, released 2026-04-20. Sparse MoE decoder, 1T total, 32B active, 256K context. Attention: MLA.
Moonshot AI, released 2026-06-12. Sparse MoE decoder, 1T total, 32B active, 256K context. Attention: MLA.
Mistral AI, released 2026-10-29. Sparse MoE decoder, 1T total, 52B active, 1M context. Attention: MLA with modified DeepSeek Sparse Attention, index reuse, and KV sharing.