Moonshot Launches Kimi K3.1 Pro: 1M-Token Context, API Prices Cut by a Third

Moonshot AI released Kimi K3.1 Pro on October 23, and you can use it today in the Kimi apps, on kimi.com and through the Kimi API. Three things matter if you use AI tools: a one-million-token context window, API prices about a third lower than the previous generation, and a promise to publish the full weights by November 4. If Moonshot follows through, anyone will be able to download and run an open model that sits close to the best closed ones. That makes this more than another version bump.

What changed

K3.1 Pro is not a new architecture. In its technical blog post, Moonshot describes it as Kimi K3 with a much larger round of post-training, aimed at coding, 3D creation, financial research and agent tasks. The size is unchanged. The Decoder reports 2.8 trillion total parameters, 104 billion active per token, and support for both text and image input.

A second variant, K3.1 Pro Highspeed, has the same capabilities at what Moonshot says is two to three times the speed, and costs more. On the practical side, the model works with the Unity, Unreal and Godot game engines and can write and run Blender and CAD scripts. Kimi.com also gets a “3D Make” feature for previewing, editing and exporting 3D models.

Moonshot’s own tables put the model close to Claude Opus 5.5 on agentic and coding work, and The Decoder notes that the independent Artificial Analysis index ranks it fourth overall. Those are mostly English-language tests, though, so we’ve included a quick check you can run yourself at the end.

Where to get it and what it costs

K3.1 Pro is live on kimi.com and the mobile apps (iOS, Android and HarmonyOS). It’s also in the Kimi Work desktop app (version 3.1.0 or later, for Windows and Apple silicon Macs) and in Kimi Code, where you select it with the /model command. On enterprise accounts it is off by default until an admin turns it on.

Kimi’s membership page, as it loaded for us, lists a free Adagio tier, then Plus at $15 a month billed annually ($180 a year), Pro at $31 ($372 a year), Max at $79 and Ultra at $159. The Highspeed variant is listed explicitly from Pro upward. What the page doesn’t say is whether free users get K3.1 Pro itself or how much of it, so test the free tier before you upgrade.

For developers, the API pricing page lists kimi-k3.1-pro at $2 per million input tokens, $10 per million output tokens and $0.20 for cached input. K3 cost $3 and $15. Highspeed costs double: $4 in and $20 out. Billing is pay-as-you-go with no subscription, and tax is added based on where you are. The API accepts both OpenAI- and Anthropic-style requests, so tools like Claude Code and Codex can point at it with a new base URL and key.

Heads-up for K2 users

The same pricing page says K2-series models will be closed to newly registered users from October 31, and all of them will be retired by the end of December 2026. If you built something on K2.6 or K2.7 Code, start planning the move now. K3.1 Pro costs more per token than those models.

The Arabic angle

Moonshot’s announcement says nothing about Arabic and doesn’t publish a list of supported languages. Prices are in US dollars, and we saw no regional pricing. In our view, the open weights matter most for the region. Banks, ministries and hospitals in the Gulf, Egypt and North Africa often have to keep data inside the country. Once the weights are out, they could host a near-frontier model on their own servers. At this size that’s a data-center project, not something for a home PC.

Before you rely on it for Arabic work, try three things:

• Write a request in a dialect (Egyptian, Gulf or Levantine) and ask for an answer in Modern Standard Arabic, then in the same dialect. Check whether it gets the meaning or translates word for word.
• Upload a long or scanned Arabic PDF and ask about a detail buried in the middle. This tests both the long context and how well it reads Arabic from images.
• Ask it to build an Arabic web page or slide deck, then check right-to-left layout, number order and punctuation.

Bottom line

K3.1 Pro makes a near-top model cheaper and promises to open it up. If you already use Kimi, it’s available now. We’ll hold the final verdict for two things: weights that actually ship on time, and real-world testing in Arabic.