# Moonshot Launches Kimi K3.1 Pro, Free to Try
> Moonshot released Kimi K3.1 Pro, which it says is much better at building apps and games, making 3D models and finance work, along with a 3D Make tool that turns a sketch or photo into a printable model. Anyone can try it free with limits, and paid plans start at $15 a month.
By Pat Simmons · October 25, 2026 · 6 min
Canonical: https://www.aiformortals.co/blog/kimi-k3-1-pro
Moonshot, the Chinese company behind the Kimi chatbot, [released Kimi K3.1 Pro](https://www.kimi.ai/blog/kimi-k3-1-pro), its most capable model yet.
Anyone can try it with limited free use on kimi.ai and in Kimi Work, Moonshot's desktop app for Windows and Apple-silicon Macs, and it's also in the Kimi app for iPhone and Android. Moonshot also says it will release the model itself as open weights by November 4, meaning the files anyone can download and run on their own computers.
Here's Moonshot's launch post:
## What Kimi K3.1 Pro does better than July's Kimi K3
K3.1 Pro is the same Kimi K3 model that came out in July, trained much further. Moonshot says that extra training made it much better at three kinds of work.
The first is **building whole apps and games**. Moonshot says it can build a complete game in Unity, Unreal and Godot, the tools studios use to make video games, and its web pages and 3D scenes come out with better layout, type and animation.
On a test of how well AI handles coding jobs on its own, Moonshot puts K3.1 Pro just behind Claude Opus 5.5 and ahead of GPT-6 Astra.
The second is **3D modeling**. From a written brief, a sketch or a reference photo, it works out the shape and proportions of an object and writes the scripts that build it in tools like Blender and CAD software, the programs designers and engineers use for 3D parts.
The third is **finance work**, like building financial models and sourced research. Moonshot says it made far fewer factual mistakes, cutting the share of made-up claims by about two-thirds on its own test.
## Turn a description, sketch or photo into a 3D model with 3D Make
Alongside the model, Moonshot added **3D Make** to kimi.ai and the Kimi app. You describe an object, upload a sketch or drop in a photo, and it builds a 3D model you can preview and edit right there.
When it's done, you can download the model ready for a 3D printer, or export it as a video or a Live Photo to share. Moonshot hasn't said whether 3D Make counts against the free plan's limits.
A 3D model from Moonshot's K3.1 Pro launch video. From Moonshot on X
A first try in 3D Make can be as simple as this:
> Make a 3D model of a desk organizer with three pen slots, a slot for a phone standing up, and rounded edges. Size it to print on a home 3D printer.
Or start from a picture you already have:
> Here's a photo of my grandmother's old teapot. Turn it into a 3D model I can print at about half its real size.
## What it costs, and why Moonshot gave subscribers a week back
The free plan gives you limited use of K3.1 Pro on kimi.ai and in Kimi Work. It doesn't include Kimi Code, Moonshot's coding tool, or K3.1 Pro Highspeed, a version Moonshot says runs two to three times faster.
Paid plans, priced per month when you pay for a year, according to [Moonshot's pricing page](https://www.kimi.ai/resources/kimi-k3-1-pro-pricing):
- **Plus, $15 a month:** adds Kimi Code and more room for the longer jobs Kimi works through on its own.
- **Pro, $31 a month:** adds K3.1 Pro Highspeed.
- **Max, $79 a month, and Ultra, $159 a month:** much higher usage allowances and very long chats.
In the first few days, those plans used up people's allowances faster than they should have, and on Sunday Moonshot [apologized on X](https://x.com/Kimi_Moonshot/status/2114450610724708492): "We're sorry for the rate limits and unnecessary quota consumption on the K3.1 Pro subscription plan. These were our issues, and we should have fixed them sooner."
Every subscriber on every plan got a one-time reset that gives back 25% of the monthly allowance, which Moonshot says equals one week of use. If you'd used less than that, your usage went back to zero. Moonshot also says it's cutting down on extra work the model did behind the scenes, like starting more side tasks than a job needed and writing overly long answers, so the allowance goes further.
## Why Moonshot held back the model's hacking skills
Moonshot says it deliberately made K3.1 Pro worse at hacking before releasing it. As the model got better at coding, it got better at finding and exploiting security holes too, so Moonshot removed hacking-related material from two stages of its training: the huge pile of text it learns from early on, and the practice tasks it trains on later.
On Moonshot's own hacking test, the released model solved noticeably fewer problems than a version trained without those cuts.
[Reuters reported](https://www.reuters.com/technology/moonshot-curbs-hacking-ability-new-kimi-model-fears-grow-over-ai-powered-2026-10-24/) that Moonshot's numbers haven't been checked by anyone outside the company.
Because anyone can download and run open weights, Moonshot says it won't release one that's better at hacking than the best AI available today, and it says K3.1 Pro's hacking skills sit behind those top models.
## The paid-post claims around the launch
The launch also came with [claims, reported by VentureBeat](https://venturebeat.com/technology/moonshots-kimi-k3-1-pro-flirts-with-open-weights-leadership-amid-paid-influencer-scandal-allegations), that Kimi asked creators to post about K3.1 Pro without labeling the posts as paid.
According to VentureBeat, screenshots posted on X show what were described as the instructions for a Kimi promotion, which the poster says he turned down. They included the line "Please DO NOT put on the paid partnership label."
VentureBeat notes the document hasn't been verified, and it's unclear whether it came from Moonshot, an outside agency or someone else, or whether anyone posted under those terms. Moonshot hadn't responded to the claims as of Sunday, according to VentureBeat.
## How Kimi's last model did in my own head-to-head builds
I put [Kimi K3, the version before this one, up against Claude Fable 5 and GPT-5.6](https://www.aiformortals.co/blog/kimi-k3-fable-5-gpt-56-build-the-same-app) in July, giving all three the same builds. Kimi built 3D scenes on par with theirs for about a quarter of what Fable 5 cost to run.
Here's that test:
Since that test, Claude Opus 5.5 came out, and this month it won every 3D build in my [Gemini 4 Argon review](https://www.aiformortals.co/blog/gemini-4-argon), against Argon, Fable 5.1 and GPT-6 Astra. That's the bar K3.1 Pro's 3D work is up against.
I'm curious about K3.1 Pro, but given the claims about unlabeled paid posts, I'm trusting my own tests over anyone's posts, and I plan to put it through the same kind of 3D builds.
## Sources
- [Moonshot: Kimi K3.1 Pro tech blog](https://www.kimi.ai/blog/kimi-k3-1-pro)
- [Moonshot: Kimi K3.1 Pro pricing explained](https://www.kimi.ai/resources/kimi-k3-1-pro-pricing)
- [Kimi on X: introducing Kimi K3.1 Pro](https://x.com/Kimi_Moonshot/status/2113680701770068067)
- [Kimi on X: apology and usage reset](https://x.com/Kimi_Moonshot/status/2114450610724708492)
- [Reuters: Moonshot curbs hacking ability of new Kimi model](https://www.reuters.com/technology/moonshot-curbs-hacking-ability-new-kimi-model-fears-grow-over-ai-powered-2026-10-24/)
- [VentureBeat: Kimi K3.1 Pro amid paid influencer allegations](https://venturebeat.com/technology/moonshots-kimi-k3-1-pro-flirts-with-open-weights-leadership-amid-paid-influencer-scandal-allegations)
