We swapped our coding agent onto the Kimi K3 API and multi-file refactors started landing green on the first pass far more often.
Kimi K3 is the newest-generation Kimi chat model and successor to Kimi K2, built for agentic workflows, long-context reasoning, tool calling, and coding. On RouterBase over an OpenAI-compatible endpoint at $2.85 / $14.25 per 1M, 5% below list.
Input
Kimi K3
Hi! I'm a helpful AI assistant. What can I do for you?
Kimi K3 API - Newest-Generation Agentic Model
The Kimi K3 API runs Kimi K3, the newest-generation model in the Moonshot AI Kimi line and the successor to Kimi K2, built for agentic workflows, long-context reasoning, tool calling, and coding.
The Kimi K3 API serves Moonshot AI Kimi K3, the newest generation of the Kimi chat line and the successor to Kimi K2. The model was built for real agentic work: long-context reasoning over big inputs, dependable tool calling, multi-step planning, and coding that spans a whole repository, from generating functions and refactoring modules to tracing bugs and reviewing diffs. You reach the Kimi K3 API through RouterBase over the OpenAI chat-completions protocol, so your existing client works unchanged and one key covers the whole catalog.
Under the hood, Kimi K3 targets the work teams actually ship: assistants and chatbots, autonomous agents, research and analysis, review bots, and long-document workflows. Billing is $2.85 per 1M input tokens and $14.25 per 1M output tokens, 5% below the list rate of $3.00 in and $15.00 out, plus just $0.285 per 1M cached input, with no per-request fee. Tool calls come back in the standard OpenAI schema, so you can wire the Kimi K3 API into an agent loop or a CI job without provider-specific glue.
What the Kimi K3 API gives you
A chat model that reasons, calls tools, and ships.
Newest-generation Kimi model
The Kimi K3 API runs the newest generation of the Kimi line, the successor to Kimi K2, tuned end to end for agentic reasoning, tool use, and coding across real work.
Agentic coding loops
Plan, call tools, edit, and re-test: the Kimi K3 API returns function calls in the standard OpenAI schema, so coding agents run multi-step tasks without custom parsing.
Long coding sessions
Large files, test suites, and tool traces stay in play across a session, so the Kimi K3 API keeps its bearings deep into a repository-scale task.
Review and refactor
Diff review, bug tracing, and batch refactoring are first-class jobs for the Kimi K3 API, which was tuned on the software loop rather than general chat alone.
Structured output
The Kimi K3 API returns reliable JSON and schema-bound responses, so codemods, extraction, and pipeline steps stay easy to parse.
Streaming, per token
The Kimi K3 API streams tokens as they generate and bills per token at $2.85 in and $14.25 out, 5% below list, with no request fee.
Get going
First call to the Kimi K3 API in three steps
Wire it once, then stream.
Create a key
One RouterBase key reaches the Kimi K3 API and every other model in the catalog, so there is nothing provider-specific to set up.
Point your client
Send any OpenAI client to routerbase.com/v1 with model=moonshotai/kimi-k3 and a messages array; the Kimi K3 API answers on the same chat-completions shape.
Stream and watch usage
The model streams tokens as they generate and returns token usage per response, so cost and latency stay visible on every call to the Kimi K3 API.
Pay only for what you use
RouterBase passes through partner-tier pricing. Compared against the model's official published API rate.
Pay only for what you use
RouterBase passes through partner-tier pricing. Compared against the model's official published API rate.
| Resolution | RouterBase | Official | Save |
|---|---|---|---|
| Input (per 1M) | $2.850 | −5% | |
| Output (per 1M) | $14.250 | −5% |
One run at the default settings (Input (per 1M)) costs $2.850.
Official price = Moonshot AI list rate for Kimi K3 ($3.00 per 1M input, $15.00 per 1M output). RouterBase serves the Kimi K3 API at $2.85 / $14.25, 5% below list.
Teams building on the Kimi K3 API
Coding agents and review bots behind one call.
Tool calls come back in the standard schema, so our agent loop ran on the Kimi K3 API the same afternoon we created the key.
Review comments from the Kimi K3 API read like a careful senior engineer wrote them, and the diffs it proposes actually apply.
At $2.85 in and $14.25 out, the Kimi K3 API let us run agentic coding jobs nightly without the bill spiking.
Streaming from the Kimi K3 API is smooth, so our editor extension renders completions token by token with no buffering tricks.
We route codemods and JSON extraction through the Kimi K3 API and the structured output parses clean every run.
Long sessions hold together - the Kimi K3 API kept a whole service plus its tests in play through a two-hour refactor.
Moving from another provider to the Kimi K3 API was one string change; the OpenAI shape did not move at all.
Our PR bot standardized on the Kimi K3 API because the review quality held steady as we scaled it across repos.
Kimi K3 API - common questions
What to know before you route traffic to it.
The Kimi K3 API is RouterBase hosted access to Moonshot AI Kimi K3, the next-generation coding model in the Kimi family, over an OpenAI-compatible endpoint.