Kimi K3.1 Pro is out: a 2.8T open-weight contender, with the weights due by Nov 4

On October 23 Moonshot AI launched Kimi K3.1 Pro, which it calls "our most capable 3T-class open-weight model." It's the first model in a new K3.1 series, and Moonshot describes it as a post-trained version of Kimi K3, its open model from this summer, trained with "substantially larger post-training compute."

What it is. A mixture-of-experts model with 2.8 trillion total parameters and 104 billion active per token, a 1-million-token context window, and image input as well as text. ("3T-class" is Moonshot's rounding; the launch post gives the exact figure, 2.8T, which matches the published Kimi K3 model card. Some third-party listings say "3T total," but that's the rounded label, not the spec.) There's also a K3.1 Pro Highspeed version that Moonshot says runs 2-3× faster.

What Moonshot says it's good at. Three things stand out in the launch materials. Full-project coding, including building games in Unreal Engine, Unity and Godot. Parametric 3D modeling, where it writes and runs scripts for Blender and CAD tools. And finance work: Moonshot says its claim-level hallucination rate on an internal stress test fell from 23.4% for K3 to 8.3%, and notes those rates "do not represent overall error rates in daily use."

The benchmarks, as Moonshot reports them. On Terminal-Bench 4.0, Moonshot says K3.1 Pro scores 65.7%, ahead of GPT-6 Astra (58.2%) and just behind Claude Opus 5.5 (66.4%). On BenchCAD with tools it reports 92.7%, against 96.2% for Claude Opus 5.5 and 95.9% for GPT-6 Astra. It also claims a state-of-the-art 64.4 on OfficeQA Pro v2. These are vendor numbers, run with Moonshot's own harness choices; we haven't seen independent results yet.

Where to use it. It's live now on Kimi.ai and the Kimi mobile apps, the Kimi Work desktop app (version 3.1.0 or later), Kimi Code (pick it with /model), and the Kimi API as kimi-k3.1-pro and kimi-k3.1-pro-highspeed. Moonshot's API platform lists K3.1 Pro at $2.00 per million input tokens and $10.00 per million output tokens ($0.20 for cache hits), and Highspeed at $4.00 and $20.00. The API docs say the model unlocks after a minimum $1 top-up. Kimi Enterprise admins have to switch it on; it's off by default.

Who it's for. Developers who want a frontier-class model for long coding and agent jobs at a lower per-token price than the big proprietary models, teams doing 3D or game work who want a model that drives Blender and game engines directly, and anyone who wants the option to self-host a top model once the weights are out.

The honest caveat. "Open-weight" is a promise today, not a download. Moonshot says the full weights will be released "by Nov 4," and when we checked Moonshot's Hugging Face page on October 23 there was no K3.1 repository yet, so the K3.1 license hasn't been published. For reference, the existing Kimi K3 License is permissive but not unconditional: a company running a model-as-a-service business with more than $20 million in revenue over 12 months needs a separate agreement before commercial use, and products with over 100 million monthly users or $20 million a month in revenue must display "Kimi K3" prominently. Expect K3.1 to look similar, but check the actual file when it ships. And open weights don't mean cheap to run: the vLLM serving recipe for the same-size Kimi K3 calls for at least eight NVIDIA GB300 GPUs, with multi-node setups for real production traffic. For most people, the API is the practical way in.

MODEL has not tested Kimi K3.1 Pro yet. Everything above comes from Moonshot's launch post on X, its tech blog, its API platform and docs, the Kimi K3 model card and license on Hugging Face, and vLLM's Kimi K3 recipe, all read on October 23. This is not sponsored and has no affiliate links.