Kimi K3 API at 100 tok/s: The Fastest Endpoint

The Fastest Kimi K3 API

Kimi K3 tops the Arena.ai Frontend Code Arena at 1,679 Elo, ahead of Claude Fable 5, the first open model to lead frontend coding. It also runs slow everywhere. Moonshot's own API clocks around 24 tokens per second under launch load with roughly 6 seconds to first token, and because the weights are not out yet, no third party can serve it faster. That is the gap Morph fills: the Kimi K3 API at about 100 tokens per second, OpenAI-compatible, with no 429 death-loops and no user-agent whitelist. Get a key and point your editor at it.

The Speed Gap

The model is not the bottleneck. The serving is. Vibe coding is an iteration loop, and at 24 tokens per second the loop breaks.

Moonshot throughput per launch-week measurements (Artificial Analysis clocks ~62 tok/s clean, still below the median for its price tier). Third-party hosts cannot serve K3 until the weights ship.

Call It From Your Editor

The endpoint is OpenAI-compatible. Set the base URL to Morph, pass your key, call the model. The same key works in any client that speaks the OpenAI API: Cline, OpenCode, Roo Code, and Cursor's custom-model setting.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.morphllm.com/v1",
    api_key="YOUR_MORPH_API_KEY",
)

resp = client.chat.completions.create(
    model="kimi-k3",
    messages=[
        {"role": "user", "content": "Build a landing page hero in React."},
    ],
)
print(resp.choices[0].message.content)

In an editor, set the provider base URL to https://api.morphllm.com/v1 and select the Kimi K3 model. No extra headers, no whitelisted user agent.

Why Kimi K3

For the full breakdown of the model, benchmarks, and architecture, see the Kimi K3 deep dive. For the other open models Morph serves, see Open Source Models.

Pricing

Pay per token, billed through one Morph account. Moonshot lists Kimi K3 at $3 per million input tokens and $15 per million output. Morph's per-token rate for the fast endpoint is on the pricing page, and one key covers every model in the lineup. Create a key to start.

FAQ

How fast is Morph's Kimi K3 API?

About 100 tokens per second, the fastest public Kimi K3 endpoint. Moonshot's own API runs roughly 24 tok/s under launch load with about 6 seconds to first token.

How do I get a Kimi K3 API key?

Sign up and create a key from the dashboard. Set base_url to https://api.morphllm.com/v1, pass the key, and call the model.

Can I use it in Cline, Cursor, or OpenCode?

Yes. Point any OpenAI-compatible client at https://api.morphllm.com/v1 and select the Kimi K3 model. No user-agent whitelist, unlike Moonshot's own API.