Kimi K3.1 Pro is Moonshot AI's new flagship, released on 23 October 2026. It is a post-trained successor to Kimi K3 with open weights: the same million-token context, the same multimodal input — text, images and video — and a focus on long coding tasks and multi-step agent work. The model is already live on Kumo: here is what is known about it, what it costs and how to start.
Context
1M tokens
1,048,576 tokens, up to 943,718 in one answer
Price on Kumo
$2.3 / $11.6
input / output per 1M tokens
Input
text, images, video
answers in text
Released
23 Oct 2026
on Kumo since 24 October
What Kimi K3.1 Pro is
Moonshot AI released Kimi K3 in July 2026 as a mixture-of-experts flagship. K3.1 Pro is not a new architecture but the next round of post-training on the same model: Moonshot calls it its flagship open-weight multimodal reasoning model and the successor to K3.
The model is built for work that runs over many steps: large edits across a repository, agent chains with tools, long-document analysis. Reasoning depth is set with the reasoning_effort parameter, and the provider lists tool calling and JSON-schema structured output.
Today the model is served by Moonshot AI itself through OpenRouter, with MXFP4 weights. Moonshot also runs a faster endpoint at double the price — it is used only when priority is requested explicitly, and Kumo does not use it.
Kimi K3.1 Pro vs Kimi K3
| Spec | Kimi K3 | Kimi K3.1 Pro |
|---|---|---|
| Released | July 2026 | October 2026 |
| Context | 1M tokens | 1M tokens |
| Input | text, images, video | text, images, video |
| Reasoning | yes | yes, with effort levels |
| Kumo price, input / output | $1.7 / $8.3 | $2.3 / $11.6 |
What is known about quality
There are no independent measurements of K3.1 Pro yet: Artificial Analysis and OpenRouter have not published indexes for it. The baseline is Kimi K3 — according to the Kumo benchmarks page on 23 October it scores 43.6 on the Intelligence Index, 76.2 on the Coding Index, 93.5% on GPQA Diamond and 46.9% on Humanity's Last Exam.
K3.1 Pro is a post-trained version of the same model, so the gains to expect first are in coding and agent work, which Moonshot emphasises. Once independent numbers appear, they will show up on the benchmarks page and in the model card.
What Kimi K3.1 Pro costs
| Where | Input | Output | Cache read |
|---|---|---|---|
| Moonshot AI | $2 | $10 | $0.2 |
| Kumo, paid from balance | $2.3 ($2.321) | $11.6 ($11.605) | $0.2 ($0.233) |
In Kumo token packages the model is charged with a ×4.5 multiplier — the same as Kimi K3. The model caches repeated context well: in agent work most of the input is re-reading the same repository or conversation, and that costs a tenth of regular input.
Kimi K3.1 Pro in the Kumo catalogue
The model card with its current price is below; it leads straight to setup.
What it is good for
- Long development tasks: spec-driven edits across a large repository, refactoring, migrations — the whole codebase fits into a million-token window.
- Agents with tools: multi-step chains where the model calls functions itself and checks the result.
- Documents together with images: screenshots, diagrams, scans and photos alongside text.
- Analysis of long material — reports, contracts, threads — without cutting it into pieces.
What Kumo has verified
Before launch Kumo runs probe requests against every model. For Kimi K3.1 Pro, plain answers, streaming, tool calling and image input over Chat Completions are confirmed.
How to connect Kimi K3.1 Pro
- Sign up in the Kumo console and top up your balance.
- Create an API key and set a spending limit on it.
- Point any OpenAI-compatible client or SDK at the gateway https://api.kumorouter.com/v1.
- Set the model to kimi-k3-1-pro and send your first request.
curl -s https://api.kumorouter.com/v1/chat/completions \
-H "Authorization: Bearer $KUMO_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "kimi-k3-1-pro",
"messages": [
{"role": "user", "content": "Review this module and propose a refactoring plan: ..."}
],
"max_tokens": 8192
}'
Python example
The official OpenAI SDK works as is — just change the base URL and the key.
from openai import OpenAI
client = OpenAI(
base_url="https://api.kumorouter.com/v1",
api_key="KUMO_API_KEY",
)
stream = client.chat.completions.create(
model="kimi-k3-1-pro",
messages=[{"role": "user", "content": "Find why this test fails from the log: ..."}],
reasoning_effort="high",
stream=True,
)
for chunk in stream:
print(chunk.choices[0].delta.content or "", end="")
FAQ
- How does Kimi K3.1 Pro differ from Kimi K3?
- It is a post-trained version of the same model: the context and input types are the same, Moonshot focuses it on long coding and agent work, and reasoning depth is adjustable.
- Does Kimi K3.1 Pro work in Russia?
- Through Kumo, yes: the gateway works without a VPN, and the balance can be topped up with a Russian card, SBP or crypto.
- Can I send images to the model?
- Yes, image input through the Kumo gateway is confirmed by probe requests. Video is listed by the provider but not yet verified on Kumo.
- Are there independent benchmarks?
- Not yet. As soon as Artificial Analysis or OpenRouter publish them, the numbers will appear on the Kumo benchmarks page.
Sources
- Model description, prices and parameters: Kimi K3.1 Pro on OpenRouter, 24 October 2026.
- Kimi K3 benchmarks: Kumo benchmarks page — Artificial Analysis data, 23 October 2026.
- Kumo prices and specs: models page, price list v41.