Short answer: “Kimi K2” can refer to the original model or newer members of the K2 family. For a new integration, distinguish older preview IDs from current model IDs: Moonshot’s changelog says the legacy Kimi K2 series was retired. Current Kimi API materials describe Kimi K2.6 and Kimi K2.7 Code as separate options. Check the model-list endpoint for availability to your account before building around either one. (Changelog; model list)
Product and API information checked November 9, 2026. Names, availability, and prices can change.
What does Kimi K2 refer to?
The original Kimi K2 was a mixture-of-experts model with 1 trillion total parameters and 32 billion activated parameters. Moonshot’s original model materials distinguish Base and Instruct variants. These refer to the original release; they should not be confused with later API model IDs. (MoonshotAI/Kimi-K2)
Moonshot’s API materials list newer K2-family IDs, including kimi-k2.6 and kimi-k2.7-code. Its changelog says several older K2 IDs—including kimi-k2-0711-preview, kimi-k2-0905-preview, and K2 Thinking and Turbo previews—were retired. Treat old tutorials and copied preview IDs as historical unless the current model list confirms availability. (API model list; changelog)
Which Kimi K2 model fits your workload?
| Model | Consider it for | Published details to weigh |
|---|---|---|
| Kimi K2.6 | General-purpose work, including coding, tool-assisted tasks, and image or video inputs | Moonshot describes K2.6 as a general-purpose model supporting text, image, and video input, with a 256K-token context window. (K2.6 quickstart) |
| Kimi K2.7 Code | Coding-focused work, especially tasks involving extended software-engineering workflows | Moonshot provides a dedicated K2.7 Code quickstart and identifies a high-speed variant. Check its documented model settings and availability before use. (K2.7 Code quickstart) |
| Original K2 preview IDs | Existing integrations that still depend on them | Moonshot’s changelog reports retirement of the listed legacy K2 series. Check model availability and migration guidance rather than assuming a preview ID still works. (Changelog) |
Product descriptions can narrow your shortlist, but they don’t establish which model will work best for your application. Test representative inputs and score correctness, completeness, tool-call validity, latency, and cost. For coding, use real repository tasks and run your tests. For multimodal work, test the image or video inputs your users will actually submit.
Production API: a minimal request and integration checks
The Kimi API documents an OpenAI-compatible Chat Completions route at https://api.moonshot.ai/v1/chat/completions. Requests use a bearer API key. The example below uses the documented K2.6 model ID and Chat Completions format; it is an example, not a claim that the request was executed. First confirm that kimi-k2.6 appears in the model list for your account. (API overview; Chat Completions API; model list)
Set MOONSHOT_API_KEY in your environment, then send a request with curl and jq installed:
curl --fail-with-body --silent --show-error \
https://api.moonshot.ai/v1/chat/completions \
-H "Authorization: Bearer $MOONSHOT_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "kimi-k2.6",
"messages": [
{"role": "user", "content": "Summarize the purpose of an API health check in one sentence."}
]
}' | jq -r '.choices[0].message.content'
The command prints the assistant’s message content from the Chat Completions response. Keep the API key out of client-side code, public repositories, and logs. If the model ID is absent from your account’s model list, don’t assume this request is available to your account; check the provider’s current documentation or support route. (API overview; Chat Completions API; model list)
Before production rollout, also verify the selected model’s supported parameters, any account or regional requirements that apply to your route, and the rate limits and error-handling behavior relevant to your expected traffic. Test tool-calling or multimodal flows separately: a model capability description does not by itself implement your application’s tool-execution loop.
Pricing: verify the applicable rate before budgeting
I can’t substantiate a usable current token-rate quote here for Kimi K2.6 or K2.7 Code—including the full rate-column mapping, currency, cache-billing conditions, and account applicability. The source material available for this guide is not enough to support a reliable cost estimate, so no numerical estimate is provided. Check the current Kimi API pricing documentation for the exact model and account route before forecasting spend. Do not substitute an older article’s K2 prices for a current quote.
When you compare API use with self-hosting, treat them as different cost bases. An API estimate needs the applicable input and output rates and realistic token usage. A self-hosting comparison also needs verified license terms, hardware requirements, and operating costs for the specific model.
Recommendation
- Building a general-purpose or multimodal feature? Start by evaluating Kimi K2.6 on representative prompts and inputs.
- Building a coding workflow? Evaluate Kimi K2.7 Code and compare its documented variants against your latency and cost requirements.
- Maintaining an older K2 integration? Check the current model list and changelog; don’t assume a legacy preview ID remains supported.
- Planning a production budget? Confirm account access and current rates for the exact model, then measure realistic token use in a pilot.
Choose based on verified model availability, workload-specific results, integration requirements, and a cost estimate grounded in the applicable provider rates—not on the K2 name alone.
Sources
Platform Changelog - Kimi API Platform
List Models - Kimi API Platform
API Overview - Kimi API Platform
Chat Completions API - Kimi API Platform