Kimi K3.1 Pro vs Kimi K3: Pricing and Upgrade Considerations

Moonshot reports 61.2% for K3.1 Pro and 55.95% for K3 on its internal long-horizon coding test. It also reports higher K3.1 Pro scores on two separate financial tasks; those results do not predict performance on every project.

Sources: ³

Vendor-reported results. Evaluation settings and sources are described in the methodology.

Additional task evidence reported by Moonshot

One financial test scores spreadsheet modeling. The other checks whether financial research is backed by evidence and can be reviewed under the test’s stated conditions. Higher scores do not guarantee financial accuracy. The release’s 3D examples have no paired K3 score, so they do not measure a 3D upgrade.³

Changing the model ID does not guarantee that existing prompts, tool calls and output formats will behave the same way. Try K3.1 Pro on jobs your K3 application already handles, including examples that previously failed.

Budget for changes to prompts or response handling as well as the token savings. The published scores do not tell you how long those changes will take or how quickly your own tasks will finish.

Sources: ¹²³

Sources: ¹²

The cold-cache example uses the published 5-minute write rates. The new prefix is charged once as a cache write and must be cache-eligible. The warm-cache example assumes a later request successfully reads an existing entry.

Each row fixes the input and output amounts to show token charges. Cache-read estimates require a successful hit. Tools, Sandbox runtime, storage, networking and taxes are separate; negotiated discounts and batch processing are excluded.

For both Kimi endpoints, Novita’s public catalog lists 1,048,576 tokens for context and maximum output; the model pages abbreviate these limits as 1M.¹⁷

Context and maximum-output values are the providers’ published limits, not a promise that both can be fully consumed in one request. Apply the endpoint’s total-context and generation rules.

Sources: ¹²⁷

Start with a small share of real jobs and keep a way to return to K3. Compare the money spent on every attempt needed to get a usable result; lower rates may be offset by extra attempts or migration work.

If the upgrade involves code changes, an optional Novita Agent Sandbox can run the same files and tests for K3 and K3.1 Pro. That helps separate model changes from changes to the environment. Your application sets tool permissions, and Sandbox usage is charged separately from model tokens.

Explore Novita Agent Sandbox · Build a repeatable execution environment

Sources: ⁵⁶

This example calls K3 and K3.1 Pro with the same planning prompt and prints the stop reason and usage for each. It is a starting point for trying the new model ID; it does not execute the proposed steps.

Generation is capped at 1,024 tokens, including any reasoning tokens counted by the endpoint. For an empty answer or a completion stopped by the token limit, the code prints a notice and does not automatically request another completion. Review usage before raising the limit.

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["NOVITA_API_KEY"],
    base_url="https://api.novita.ai/openai/v1",
)

for model in ["moonshotai/kimi-k3.1-pro","moonshotai/kimi-k3"]:
    reply = client.chat.completions.create(
        model=model,
        messages=[{"role": "user", "content": "Review a repository bug fix: list the implementation steps and acceptance tests."}],
        max_tokens=1024,
    )
    choice = reply.choices[0]
    content = choice.message.content or ""
    print(model, "finish_reason:", choice.finish_reason)
    print("usage:", reply.usage)
    if choice.finish_reason == "length" or not content.strip():
        print("No complete answer. Inspect usage and adjust the generation budget before another request.")
    else:
        print(content)

Sources: ⁴

Sources: ¹²³