Moonshot AI's flagship Kimi K3 model launched on the Telnyx Inference API, marking the first time a 2.8-trillion-parameter open-source model has reached commercial availability through a major inference platform.

The model represents the largest open-weight system available on Telnyx's infrastructure, built on proprietary Kimi Delta Attention and Attention Residuals architectures. K3 supports a 1-million-token context window and processes text, images, and video input natively without requiring separate vision adapters.

Telnyx priced K3 at $2.70 per million input tokens and $13.50 per million output tokens, with cached input tokens at $0.27 per million. The platform automatically applies prefix caching for repeated prompt segments across requests.

Technical capabilities

K3 offers three configurable reasoning levels — low, high, and maximum — allowing developers to trade computational resources for deeper analysis per request. The model supports function calling, dynamic tool loading, and JSON schema-constrained output for agentic workflows.

The system handles codebase analysis, long document processing, and multi-turn agent sessions through its extended context window. Telnyx runs K3 on company-owned GPU infrastructure accessible via OpenAI-compatible APIs.

K3 joins existing models on the Telnyx platform including Kimi K2.6, GLM-5.2-FP8, and MiniMax M3. The model competes with closed-source frontier systems from Anthropic and OpenAI on coding and reasoning benchmarks, according to Telnyx.

The launch signals continued convergence between open-source and proprietary model performance in the trillion-parameter class. Developers can access K3 through Telnyx's Mission Control dashboard or integrate directly via the inference API documentation.