Models / Kimi K2.6
Kimi K2.6 on Gonka
Technical specifications and API capabilities of the Kimi K2.6 model deployed on the Gonka decentralized AI network.
Network model: Kimi K2.6
Architecture: Mixture of Experts (MoE)
Customer API alias:
Architecture: Mixture of Experts (MoE)
Customer API alias:
gonka-autoModel scale
- 1.06 trillion total parameters
- 32 billion active parameters per token
- 384 routed experts plus one shared expert
- Eight routed experts selected per token
- 256K native context window
What it is designed for
Kimi K2.6 combines long-context reasoning, coding, tool use, and agentic workflows. The base model is natively multimodal; the customer API exposes capabilities supported by the currently verified Gonka route.
Gonka deployment
Official Gonka node configuration deploys Kimi K2.6 with a 240,000-token operational model limit and tensor parallelism across eight GPUs. Limits exposed by Gonka Broker may be lower for service safety and cost control.
API access
Use the OpenAI-compatible POST /v1/chat/completions endpoint with the customer alias gonka-auto. Route selection remains managed by Gonka Broker.
Next step: Create an account or read the integration docs.