Kimi K2.6
Active Parameters
1T
Context Length
262K
Modality
Multimodal
Auxiliary Parameters
450M
Architecture
Mixture of Experts
KV Cache Rate
69 KB/tok
License
Modified MIT
Release Date
21 Apr 2026
Knowledge Cutoff
-
API Pricing (per 1M)
Input: $0.95 · Output: $4.00
Hardware Sizing
Precision
Context
Workload Memory: 1938.09 GB
FP16 · 8K context
Base Weights: 1922.25 GB
KV Cache: 13.34 GB
Runtime & Graphs: 2.50 GB
Minimum Required Hardware
Cluster: 16x NVIDIA H200 141GB
1938.1 / 2030.4 GB usable · TP 8 · PP 2
Determine requirements for datacenter hardware:
Upgrade to Pro
Architecture Diagram
Evaluation Benchmarks
Rank
#103
| Benchmark | Score | Rank |
|---|---|---|
Graduate-Level QA | 0.905 | 22 |
Coding | 0.79 | 31 |
Agent Arena | -0.29 | 37 |
Agentic Coding | 0.47 | 44 |
Reasoning | 0.79 | 52 |
General Text | 1461 | 52 |
Data Analysis | auto 0.65 | 53 |
General | 0.71 | 53 |
Mathematics | 0.84 | 53 |
LiveBench Average | 0.71 | 53 |
Web Development | 1509 | 54 |
Coding Index | 0.62 | 64 |
Agentic Index | auto 0.20 | 79 |
Intelligence Index | auto 0.27 Standard 0.24 | 95 124 |
Rankings
Overall Rank
#103
Coding Rank
#56
About Kimi K2.6
Kimi K2.6 is Moonshot AI's open-source native multimodal agentic model with 1T total parameters and 32B activated per token. Built on a hybrid MoE architecture with 61 layers, 384 routed experts + 1 shared, 8 selected per token, MLA attention, and a dedicated MoonViT vision encoder (400M params). Delivers state-of-the-art performance in long-horizon coding (SWE-Bench Pro 58.6%, SWE-Bench Verified 80.2%), agentic workflows (BrowseComp 83.2%, AIME 2026 96.4%, GPQA-Diamond 90.5%), and visual reasoning (MMMU-Pro 79.4%). Supports 256K native context, thinking/instant modes, and thinking preservation across turns. Scalable to 300 sub-agents executing 4,000 coordinated steps. Released April 21, 2026 under Modified MIT License.
Technical Specifications
Attention
Attention Structure
Multi-Layer Attention
Attention Heads
64
Key-Value Heads
64
Position Embedding
ROPE
RoPE Theta
50,000
Normalization
RMS Normalization
Activation Function
SwigLU
Dimensions
Auxiliary Parameters
450M
Hidden Dimension Size
7,168
Number of Layers
61
FFN Intermediate Size (Dense)
2,048
Multi-Token Prediction Heads
0
Tokenizer
Vocabulary Size
163,840
Mixture of Experts
Total Expert Parameters
32.0B
Number of Experts
384
Active Experts
9
Shared Experts
1
FFN Intermediate Size (per Expert)
2,048
Dense Layers Before MoE
1
About Kimi K2.6
Kimi K2.6 is Moonshot AI's latest open-source native multimodal agentic model, advancing practical capabilities in long-horizon coding, coding-driven design, proactive autonomous execution, and swarm-based task orchestration. It transforms simple prompts and visual inputs into production-ready interfaces and full-stack workflows, and can scale horizontally to 300 sub-agents executing 4,000 coordinated steps. Built on the same hybrid MoE architecture as Kimi K2.5 with a dedicated MoonViT vision encoder.
Other Kimi K2.6 Models
- No related models available