Actual costs depend on token usage and cache hits. Estimates use Modal’s current pricing.
Moonshot models
Serve Moonshot AI’s Kimi models on Modal for coding, reasoning, and document analysis. Access Kimi K3.1 Pro and Kimi K3 through Model APIs with per-token billing, or deploy Kimi K3 and Kimi K2.6 on Dedicated Endpoints with compute-based billing.
Which Moonshot models can you run on Modal?
Moonshot
Kimi K3.1 Pro
Moonshot AI’s multimodal model for coding, financial research, and agentic work, with a one-million-token context window.
Text VisionMoonshot AI
Kimi K3
Open frontier intelligence for deep reasoning and knowledge work.
Text VisionMoonshot AI
Kimi K2.6 NVFP4
Moonshot AI's Kimi K2.6 in NVIDIA's NVFP4 quantization: a 1T-parameter MoE agentic model with 32B active per token, 256K context, and thinking mode.
Text
| Model | Architecture | Context | Modalities | Input / Output |
|---|---|---|---|---|
| Kimi K3.1 Pro | 2.8T MoE | 1M | Text · Image | $2.00 / $10.00 |
| Kimi K3 | 2.8T MoE | 1M | Text · Image | $3.00 / $15.00 |
| Kimi K2.6 NVFP4 | 1T MoE · 32B active | 256K | Text | Dedicated only |
All prices per million tokens. Dedicated Endpoint compute is billed by the second. See compute pricing.
Frequently asked questions about Kimi models on Modal
-
Kimi models have downloadable weights, with licenses that vary by release. K3 and K3.1 Pro use their own model licenses. The K2.6 checkpoint served on Modal uses NVIDIA’s NVFP4 quantization and carries the NVIDIA Open Model License alongside upstream licensing information. See the model cards for K3.1 Pro, K3, and K2.6.
-
Yes. Use an OpenAI-compatible client with the API URL, model name, and authentication details from your Modal dashboard. The API request examples show how to send requests.
-
With a Model API, Modal manages the hardware and scaling, and you pay per token. Dedicated Endpoints provide a separate deployment with capacity controls and compute-based billing. K3 and K2.6 are available through Dedicated Endpoints.
Latest updates

Scaling Intelligence: A landmark commission and exhibition

Sandbox resourcing: burst, don't reserve
Why you're paying for capacity you aren't using, and how to pay only for what you use.

Modal becomes a gVisor maintainer
Modal runs upwards of 40 million gVisor containers every day. Here's why we bet on it, what we've built into it, and where we want to take it next.
