Kimi K3 is now available on AI Gateway

Kimi K3 from Moonshot AI is now available on AI Gateway.
K3 is an open-source model with a 1M-token context window and native visual understanding, accepting text, image, and video inputs.
Built for long-horizon software engineering, knowledge work, and deep reasoning, K3 is especially strong where code meets visual and spatial reasoning, which suits frontend, game development, and CAD workflows. Thinking mode is always on.
To use Kimi K3, set model
to moonshotai/kimi-k3
in the AI SDK:
import { streamText } from 'ai';
const result = streamText({ model: 'moonshotai/kimi-k3', prompt: 'Trace this rendering bug across the codebase and open a PR with a fix.',});
AI Gateway provides a unified API for calling models, tracking usage and cost, and configuring retries, failover, and performance optimizations for higher-than-provider uptime. It includes built-in custom reporting, Zero Data Retention support, budgets for API keys, routing rules, and more.
AI Gateway reflects provider pricing with no markup and does not charge a platform fee on inference, including on Bring Your Own Key (BYOK) requests.
Try Kimi K3 in the model playground.
AI Gateway: Track top AI models by usage
The AI Gateway model leaderboard tracks the most popular models over time, ranking them by the total volume of tokens processed across all Gateway traffic.
View the leaderboard

Facts Only

* Kimi K3 is available on AI Gateway.
* K3 is an open-source model with a 1M-token context window.
* K3 has native visual understanding, accepting text, image, and video inputs.
* K3 is built for long-horizon software engineering, knowledge work, and deep reasoning.
* K3 excels where code meets visual and spatial reasoning, suitable for frontend, game development, and CAD workflows.
* Users can call Kimi K3 using the AI SDK with the model set to moonshotai/kimi-k3.
* The AI Gateway offers unified API access, usage tracking, cost management, and performance optimizations.
* AI Gateway reflects provider pricing without markup on inference.
* The AI Gateway includes features such as custom reporting, Zero Data Retention support, and budgeting.

Executive Summary

Kimi K3 is an open-source model available on the AI Gateway, featuring a 1M-token context window and native visual understanding, capable of processing text, image, and video inputs. It is designed for complex tasks like long-horizon software engineering, knowledge work, and deep reasoning, showing particular strength in areas involving code and visual/spatial reasoning relevant to frontend, game development, and CAD workflows. Users can access Kimi K3 via the AI SDK by setting the model to moonshotai/kimi-k3. The AI Gateway provides a unified API for model calls, usage tracking, cost management, and performance optimizations, including features like custom reporting and Zero Data Retention support. Furthermore, the AI Gateway reflects provider pricing without markup on inference requests.

Full Take

The integration of a model with native visual understanding and a large context window into a platform focused on enterprise-grade API management suggests a shift toward multimodal reasoning being commoditized and standardized through infrastructure layers like AI Gateway. The emphasis on K3’s application in spatial reasoning connects theoretical model capabilities directly to tangible, high-stakes engineering domains like CAD and game development, implying that the value proposition is moving from pure language processing toward embodied, contextual understanding. The existence of a centralized gateway managing usage, cost transparency, and uptime suggests an attempt to introduce governance and reliability into the decentralized landscape of open-source model access. This dynamic raises questions about where the responsibility for visual grounding resides—with the model architecture, the context provided by the input modalities, or the framework managing the reasoning steps. What measures are in place to ensure that this multimodal capability translates reliably from theoretical reasoning strengths (as noted by "Thinking mode is always on") into verifiable, safe outputs within these highly specialized fields?

Sentinel — Human

Confidence

The text reads like a direct product announcement or feature walkthrough, combining factual specifications with promotional framing, which is typical of human-authored technical marketing copy.

Signals Detected

Moderate sentence length variance; direct, promotional tone inconsistent with typical high-level academic prose.

Clear, functional flow focused on feature announcement and API integration; lacks the hedging or expansive contextualization common in purely synthetic content.

Direct instructional code snippets and specific product names suggest direct reporting rather than templated argument.

Claims regarding model features (1M-token context window, visual understanding) are specific and appear grounded in the context of a product announcement.

Human Indicators

The inclusion of specific technical implementation details (SDK usage, API calls) mixed with marketing language suggests an origin from a developer blog or press release.

The functional flow moves smoothly between introducing the model's capabilities, showing usage instructions, and describing the platform infrastructure.