Kimi API access to Moonshot AI's Kimi models on APIMart: Kimi K3 with a 1M-token context and native vision, K2.7 Code and K2.7 Code Highspeed for programming, K2.6 with thinking and non-thinking modes, and K2.5 for existing workflows.
Guides, comparisons and pricing analysis from the APIMart blog.
What's on your mind?
Ask APIMart
kimi-k3
Transparent pricing with no hidden fees. Pay only for what you use.
Per 1M tokens
| Rate | GoldCurrent price | Platinum | Diamond | Official | Your savings |
|---|---|---|---|---|---|
| Input | 24~$2.4 | 22.8~$2.28 | 21.6~$2.16 | 30~$3 | 20% |
| Cached input | 2.4~$0.24 | 2.28~$0.228 | 2.16~$0.216 | 3~$0.3 | 20% |
| Output | 120~$12 | 114~$11.4 | 108~$10.8 | 150~$15 | 20% |
Call Moonshot AI's Kimi models through one OpenAI-compatible API on APIMart. Kimi K3 is the flagship for long-context reasoning and agentic work, K2.7 Code is built for programming, and K2.6 handles dialogue and agent tasks with optional thinking.
50K+
Active Users
99.9%
Uptime
2x
Faster
70%
Cost Savings
What each Kimi model on APIMart is built for, based on Moonshot AI's documentation
How teams use Kimi models in production
From signup to your first Kimi API call on APIMart
Create your APIMart account and top up your balance; Kimi usage is billed per token from that balance.
Generate an API key on the API Keys page. The same key works for kimi-k3, kimi-k2.7-code and every other Kimi model.
Try the playground above, or send a chat completions request to https://api.apimart.ai/v1 with the OpenAI SDK and model set to kimi-k3.
What developers say about our Kimi API
“Kimi K2's coding ability is seriously impressive. It handles complex refactoring tasks with ease.”
Alex Chen
Senior Developer
“K2-Thinking solved math problems that GPT-4o couldn't. The reasoning depth is remarkable.”
Sarah Li
AI Researcher
“K2.5 is a big step up. We've integrated it into our content pipeline for Chinese and English.”
Mike Wang
Content Lead
“Competitive pricing and solid performance. Kimi is an excellent alternative to the big providers.”
Emily Zhou
Product Manager
“The thinking mode is perfect for our legal analysis tool. It works through complex cases methodically.”
David Liu
Legal Tech Engineer
“Fast API, clean documentation, and reliable uptime. Very happy with the switch to APIMart for Kimi.”
Lisa Zhang
Backend Developer
Models, pricing, tool calling and access for the Kimi API on APIMart
The Kimi API gives you programmatic access to Moonshot AI's Kimi models. On APIMart you can call kimi-k3, kimi-k2.7-code, kimi-k2.7-code-highspeed, kimi-k2.6 and kimi-k2.5 through the same OpenAI-compatible chat completions endpoint and one API key.
Kimi K3 is Moonshot AI's 2.8-trillion-parameter model, released on July 16, 2026. It has a 1M-token context window, native vision and always-on reasoning, and Moonshot AI also published open weights. It is built for long-horizon coding and deep reasoning.
Kimi K3 is billed per million tokens. APIMart's list price matches Moonshot AI's official rates ($3.00 input, $0.30 cached input, $15.00 output); Gold members pay 20% less ($2.40, $0.24 and $12.00, October 2026), and Platinum and Diamond members save up to 28%. Failed requests are not charged. The table above shows live prices.
You pay for input tokens and output tokens separately, at the per-million-token rate of the model you call. On Kimi K3, cached input tokens are billed at the lower cached-input rate. Check the pricing table above for each model's current rates.
Choose Kimi K3 for the strongest reasoning, long context and agentic work. Use K2.7 Code for programming, or K2.7 Code Highspeed when output speed matters. Use K2.6 for dialogue and agent tasks where you want to switch thinking off for faster answers, and K2.5 only for existing workflows built on it.
Kimi K3 has a 1M-token context and always reasons, with low, high or max effort. Kimi K2.6 has a 256K context and lets you turn thinking on or off; Moonshot AI fixes its temperature at 1.0 in thinking mode and 0.6 in non-thinking mode. Moving from K2.6 to K3 is mostly a model ID change plus removing the K2 thinking settings.
Yes. Kimi models support tool calling with the standard tools field. On Kimi K3, tool_choice can be auto, none or required; K2.6 and K2.7 Code do not support required. In multi-step tool loops, send the full assistant message back, including reasoning_content.
Yes. Point the OpenAI SDK at https://api.apimart.ai/v1, use your APIMart API key, and set model to a Kimi model ID such as kimi-k3. Requests use the standard chat completions format.
Moonshot AI discontinued kimi-k2.5 on its own platform on August 31, 2026 and plans to retire all K2 series models by the end of December 2026. The K2 models listed on this page are still available on APIMart; for new projects, Kimi K3 is the recommended choice.
Sign up for APIMart, open the API Keys page in your dashboard and create a key. The same key works for every Kimi model and every other model on the platform, with pay-as-you-go billing and no subscription.
No. There is no free quota for the Kimi API on APIMart. Billing is pay-as-you-go: failed requests are not charged, and members get 20% or more off every price.
No. Under APIMart's Terms of Service, your data is not used to train AI models or shared with third parties except as needed to provide the service. Requests are processed only to serve your API calls.
Explore more models in the same category.
Gemini 3.5 Flash Lite
A lightweight multimodal model launched by Google that emphasizes low cost, low latency, and high throughput. It is suitable for document parsing, data extraction, structured output, and large-scale agent workflows, delivering faster response times while maintaining core reasoning capabilities.
Gemini 3.6 Flash
Gemini 3.6 Flash is a faster, more token-efficient, and more code-capable version of the Gemini Flash model, suitable for everyday development, AI agents, and high-concurrency applications.
Kimi K2.7 Code
A code-specific model launched by Moonshot AI that excels at code generation, understanding, debugging, and complex development tasks, and is suitable for use as an AI programming assistant.
Grok 4.5
xAI has launched a new generation of general-purpose large language models that enhance reasoning, coding, and real-time information processing capabilities, making them suitable for complex question-answering, development, and agent tasks.