Moonshot AI Kimi K3 2.8T Model Available on Telnyx Inference API
Telnyx launched hosting for Moonshot AI's flagship Kimi K3 model, featuring 2.8 trillion parameters, a 1M-token context window, and native vision. The model supports function calling and OpenAI-compatible API endpoints.
Impact: Medium
Why it matters
Developers can query a frontier 2.8T open-weights model with prompt caching and configurable reasoning effort via standard OpenAI client libraries.
TL;DR
- 01Access a 2.8T parameter model via standard OpenAI client libraries.
- 02Leverage automatic prompt prefix caching for multi-turn agent conversations.
- 03Configure reasoning depth dynamically to optimize compute budget.
Key facts
- Parameters
- 2.8 Trillion (Open-Weight)
- Context Window
- 1M Tokens
- Model ID
- moonshotai/Kimi-K3
- API Compatibility
- OpenAI REST API
Model Specifications
Kimi K3 introduces a 2.8T parameter architecture with native vision and text input capabilities, handling context lengths up to 1M tokens for codebase analysis and agent workflows.
Reasoning Modes & Function Calling
Developers can pass reasoning_effort parameters (low, high, max) to adjust inference depth per request. Built-in function calling and structured JSON output contracts allow easy integration into existing agent tools.
API Access
Requests run on Telnyx GPU infrastructure using standard OpenAI SDK setups by pointing the base URL to Telnyx and requesting model moonshotai/Kimi-K3.
Try it in 2 minutes
from openai import OpenAI
client = OpenAI(
base_url="https://api.telnyx.com/v2/ai",
api_key="TELNYX_API_KEY"
)
response = client.chat.completions.create(
model="moonshotai/Kimi-K3",
messages=[{"role": "user", "content": "Analyze codebase architecture"}]
)python
✓ When to use
- Large codebase analysis requiring up to 1M context tokens on open-weight architectures.
- Multimodal agent workflows needing function calling and structured JSON output.
✕ When NOT to use
- Low-latency real-time voice or ultra-lightweight client-side applications.
What to do today
- Test Kimi K3 endpoint on Telnyx using existing OpenAI SDK configurations.
- Configure prompt prefix caching for long agent session logs.
Sources