Online

Kimi-K2.5

Moonshot AI · OpenAI-compatible API

moonshotai/kimi-k2.5

Create account →Quickstart ↓

USD / 1M tokens

Input$0.54

Cached Input$0.06

Output$2.78

Context: 256K
Tool Calling: ✓
Reasoning: ✓
Vision: ✓
Streaming: ✓

OpenAI-compatible API

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.iteracompute.com/v1",
    api_key=os.environ["ITERACOMPUTE_API_KEY"],
)
response = client.chat.completions.create(
    model="moonshotai/kimi-k2.5",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(response.choices[0].message.content)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.iteracompute.com/v1",
  apiKey: process.env.ITERACOMPUTE_API_KEY,
});
const response = await client.chat.completions.create({
  model: "moonshotai/kimi-k2.5",
  messages: [{ role: "user", content: "Hello!" }],
});
console.log(response.choices[0].message.content);

curl https://api.iteracompute.com/v1/chat/completions \
  -H "Authorization: Bearer $ITERACOMPUTE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "moonshotai/kimi-k2.5",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

moonshotai/kimi-k2.5View Docs →

Catalog snapshot: 2026-10-16. Live prices have not been verified yet.

Overview

Kimi-K2.5 is available through the IteraCompute OpenAI-compatible API, with 256K context, streaming and reasoning.

Tasks and selection tradeoffs

Visual document experiments

Compare descriptions against labeled documents before using them automatically.

Agent workflow trials

Run a bounded task with a fixed tool budget and verify each result.

Task suggestions based on catalog features and price, not model quality benchmarks.

Choose your tool

OpenAI SDKCall from Python or JavaScript Vercel AI SDKBuild a TypeScript application

Choose a model and tool for your task →

Model capabilities

Features available through this model interface. Updated October 16, 2026.

Capability	Model developer documentation	Route test
Tool calling	Supported	Supported
Reasoning	Supported	Supported
Image input	Supported	Supported
Streaming	Supported	Supported
JSON mode (json_object)	Supported	Supported

Context: 256K · Output limit: 16K

Model developer documentation

Set thinking controls per model. Use one control style per request.

For JSON responses, validate your application schema before use. Input and requested output share the context budget.

Pricing

Prices in USD per 1M tokens.

Input$0.54

Cached Input$0.06

Output$2.78

Example task: Evaluate a bounded agent task

Assume 20,000 aggregate input tokens and 2,000 billable output tokens across all calls. No cache hit.

20,000 × $0.54 / 1M + 2,000 × $2.78 / 1M = $0.016360

If usage actually reports 15,000 of those input tokens as cached, replace that portion at the cached-input rate ($0.06/1M): $0.009160. Cached tokens are part of total input, never added on top.

Illustrative calculation using the 2026-10-16 catalog snapshot. Include every agent call and retry, repeated history and billable reasoning output; settle against actual usage and current Console prices.

Usage limits

Input and requested maximum output share the context budget. Keep both within the model limits.
Split long workflows into bounded steps and handle tool errors and retries in your application.
Reasoning tokens are billable output. Reserve enough generation budget to obtain a final text reply.
View current availability and prices in the Console.

API parameters

Parameter / protocol field	Usage
`model / messages`	Use the exact model ID and chat message list
`stream=true / stream_options.include_usage=true`	Receive SSE chunks, [DONE] and final usage
`tools / tool_calls / role=tool`	Define functions in tools, execute tool_calls and return results with role=tool
`response_format.type=json_object`	Request valid JSON; validate the required structure in your application
`reasoning_content`	Reasoning output is separate from the final reply and counts toward output usage
`{"enable_thinking":false}`	Supported; use one thinking-control style per request
`{"enable_thinking":true}`	Supported; use one thinking-control style per request

API

Base URL: https://api.iteracompute.com/v1
Endpoint: POST /v1/chat/completions
Model ID: moonshotai/kimi-k2.5
Authentication: Authorization: Bearer ITERACOMPUTE_API_KEY

View Docs

FAQ

How do I use Kimi-K2.5 with the OpenAI SDK?

Set base_url to https://api.iteracompute.com/v1 and model to moonshotai/kimi-k2.5. Authenticate with your IteraCompute API key.

What does Kimi-K2.5 cost?

See the pricing section for current input, cached-input and output rates in USD per 1M tokens. Check the refresh time before use.