Kimi K2.5

by Moonshot AI

Zero data retention−5%Retiring
moonshotai/kimi-k2.5
Input
$0.74/1M tokens
Output
$4.65/1M tokens
Context
262Ktokens
Max output
66Ktokens
Knows to
-
Released
27 Jan 2026

Scheduled for retirement on 18 Nov 2026. Requests to moonshotai/kimi-k2.5 will stop working after that date; the comparisons below are a good place to pick a successor.

About

Kimi K2.5 is Moonshot AI's native multimodal model, delivering state-of-the-art visual coding capability and a self-directed agent swarm paradigm.

Pricing

Input$0.74 from $0.69per 1M prompt tokens
Output$4.65 from $3.50per 1M generated tokens, reasoning included
Cached input$0.29per 1M tokens read back from the prompt cache

5% off right now. The provider is running a promotion, and the developer API bills what the provider actually charged for your call plus our flat markup, so it reaches your bill with nothing to claim. The “from” figure is the cheapest endpoint serving the model; which one answers is the provider’s routing decision, so treat it as a floor. Promotions can end at any time.

What a request costs

Typical requestEachPer 1,000
Short chat reply1,500 in · 400 out$0.0030$2.98
Question over a long document40,000 in · 800 out$0.034$33.52
Agent tool step20,000 in · 1,200 out$0.020$20.48

At the developer-API rate, before any provider promotion or prompt-cache saving. In the app, the model picker shows what a message costs on your plan before you send it.

Benchmarks

Coding46.8#103 of 167

Head-to-head rankings

  1. #15Godot gamesAgent arenaElo 121960% wins
  2. #29Full stackAgent arenaElo 109855% wins
  3. #32Mobile appsAgent arenaElo 114755% wins
  4. #33Android appsAgent arenaElo 111658% wins
  5. #34Web appsAgent arenaElo 112052% wins
  6. #36ASCII artModel arenaElo 116347% wins
  7. #47WebsitesModel arenaElo 124555% wins
  8. #50UI componentsModel arenaElo 123154% wins
  9. #51CodeModel arenaElo 123154% wins
  10. #523DModel arenaElo 119853% wins

Scores from Artificial Analysis; bars are scaled to the best model on our list. Rankings from Design Arena head-to-head votes. Published by OpenRouter alongside its catalogue.

Call it from code

OpenAI-compatible: point any OpenAI SDK at https://api.privateer.pro/v1 and set model. Keys are created in the app under Settings.

cURL
curl https://api.privateer.pro/v1/chat/completions \
  -H "Authorization: Bearer $PRIVATEER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "moonshotai/kimi-k2.5",
    "messages": [{ "role": "user", "content": "Hello" }]
  }'
Python
import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.privateer.pro/v1",
    api_key=os.environ["PRIVATEER_API_KEY"],
)
reply = client.chat.completions.create(
    model="moonshotai/kimi-k2.5",
    messages=[{"role": "user", "content": "Hello"}],
)
print(reply.choices[0].message.content)
JavaScript
import OpenAI from "openai";

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

Parameters this model accepts

frequency_penaltyinclude_reasoninglogit_biaslogprobsmax_tokensmin_ppresence_penaltyreasoningrepetition_penaltyresponse_formatseedstopstructured_outputstemperaturetool_choicetoolstop_ktop_logprobstop_p

API reference → · How API billing works →

Prices are USD at the developer-API rate, as of 9 Nov 2026, rebuilt daily from the providers’ live catalogues. ← All models