Modal offers Kimi K3.1 Pro on Shared Endpoints with token billing

Modal says Kimi K3.1 Pro, a post-training update to Moonshot AI's Kimi K3, is now available on its Shared Endpoints, billed per token through an OpenAI-compatible API. Modal says the model supports text and image inputs with a 1M-token context window and was updated with expanded reinforcement learning for coding, financial research, and long-horizon agentic work.

Modal says it worked with SGLang on the model bringup, validating checkpoint accuracy and optimizing the serving path for day-zero support. SGLang says Kimi K3.1 Pro replaces dense MLA with sparse attention (DSA) and reports up to 33.7K tokens/s per GPU on GB300 and 450 tokens/s for a single user using DFlash and MegaKernel. These performance figures come from SGLang and have not been independently verified.