Home › Blog › Kimi API Guide

Kimi API Guide 2026: Moonshot K3 Pricing & 256K-Context Workloads

Kimi K3's 256K context window makes it a natural fit for long-document agents — but official pricing (¥21.9 input / ¥109.5 output per 1M tokens) is the steepest of the major Chinese labs. Bridge serves K3 at $2.47/$12.34 — 20% below official.

Kimi K3 pricing: official vs Bridge

Cheaper Kimi tiers. For long-context or coding workloads that don't need K3's flagship reasoning, Kimi K2.6 and K2.7 Code are on Bridge at $0.56/$1.13 per 1M tokens — same OpenAI-compatible endpoint, swap by changing one string.

When to choose Kimi K3

• Long-document analysis — contracts, research papers, codebases in a single context window.
• Agentic workflows — multi-step tool use with long histories; cached-input billing keeps repeated context cheap.
• Deep reasoning on big state — where the 256K window beats chunking workarounds.

Python example

from openai import OpenAI

client = OpenAI(
    base_url="https://bridge.computeindex.net/v1",
    api_key="sk-bridge-YOUR_KEY",
)

completion = client.chat.completions.create(
    model="kimi-k3",
    messages=[
        {"role": "system", "content": "You summarize long legal documents."},
        {"role": "user", "content": open("contract.txt").read()},
    ],
    stream=True,
)
for chunk in completion:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")

Long context, predictable bill

256K-context workloads are exactly where token accounting matters most: a single 200K-token prompt at official output prices gets expensive fast. Bridge meters every request from the upstream usage field, applies the cached-input discount automatically, and lets you reconcile any request by its x-request-id.

FAQ

Is Kimi K3 on Bridge the same as Moonshot's official model? Yes — same model, served through the OpenAI-compatible Bridge gateway.

Can I stream K3 responses? Yes — standard SSE streaming with stream_options.include_usage supported.