Kimi K2.5

Moonshot’s multimodal agent model for long-context planning, browsing, and product workflows.

multimodal agentPublic model detailMoE Transformer

Params

1T / 32B active

Context

256K

Max Output

64K

License

Modified MIT

TTFT

520ms

Throughput

42 tok/s

Why pick it

Huge context for agent memory
Balanced multimodal reasoning

Pricing

TierStandardCachedSiliconFlowSavings

Realtime$0.10 / $1.00$0.035$0.23 / $3.0057%

Batch$0.05 / $0.50$0.035$0.23 / $3.0057%

Quick start

OpenAI-compatible surface. Swap the base URL and ship.

Try in Playground Open pricing

Python

from openai import OpenAI

client = OpenAI(
    base_url="https://api.luminapath.tech/v1",
    api_key="BATCHIN_API_KEY"
)

resp = client.chat.completions.create(
    model="kimi-k2.5",
    messages=[{"role": "user", "content": "Summarize why this model is a fit for my workload."}]
)

print(resp.choices[0].message.content)

JavaScript

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.luminapath.tech/v1",
  apiKey: process.env.BATCHIN_API_KEY,
});

const resp = await client.chat.completions.create({
  model: "kimi-k2.5",
  messages: [{ role: "user", content: "Summarize why this model is a fit for my workload." }],
});

console.log(resp.choices[0]?.message?.content);

cURL

curl https://api.luminapath.tech/v1/chat/completions \
  -H "Authorization: Bearer $BATCHIN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kimi-k2.5",
    "messages": [{"role":"user","content":"Summarize why this model is a fit for my workload."}]
  }'

Specs

Architecture

MoE Transformer

Vendor group

Moonshot

Context window

256K

Max output

64K

Best for

agent

multimodal