Moonshot AI's post-trained upgrade to Kimi K3 and the first model in its K3.1 series — a 3T-class open-weight model with a 1M-token context, strong coding and agentic work, lower hallucination, and a 2–3x-faster Highspeed variant.
Last verified · 2026-10-23 · by Moe Ameen
Kimi K3.1 Pro is the first model in Moonshot AI's new K3.1 series, launched in late October 2026 by the Beijing lab behind the Kimi series. Moonshot describes it as a post-trained version of its Kimi K3 flagship — the same base model trained with substantially more post-training compute — and calls it its most capable "3T-class" open-weight model to date. It is served across kimi.ai, the Kimi apps, Kimi Work, Kimi Code, and the Kimi API, where the model IDs are kimi-k3.1-pro and kimi-k3.1-pro-highspeed. For organizations it is off by default until an administrator enables it, and Zero Data Retention is available to API customers.
It keeps K3's headline features: a 1-million-token context window and native multimodal understanding, so images and screenshots are first-class input alongside text. Moonshot describes the scale only as "3T-class"; third-party write-ups put it around 2.8 trillion total parameters with a fraction active per token, so treat the exact count as approximate until the model card is public. A faster sibling, Kimi K3.1 Pro Highspeed, ships alongside it — Moonshot says it runs roughly two to three times faster with identical capability, at (per third-party listings) about double the price and quota. Full open weights are slated to follow the hosted launch, with Moonshot pointing to on or around November 4, 2026.
Rather than one headline number, Moonshot frames K3.1 Pro around coding, agentic work, 3D creation and modeling, and financial research. Its own benchmarks put coding near Claude Opus 5.5 (a Terminal-Bench 4.0 figure around 65.7% versus 66.4%) and claim a sharp drop in hallucination versus K3 on its internal stress test — a reliability gain that matters for research-heavy work. Those are first-party benchmarks, so wait for independent leaderboards before treating any single figure as settled. Reported launch API pricing is near $2 per million input tokens and $10 per million output (cached input around $0.20), below K3's $3/$15, though the figures come from third-party providers.
The honest framing: K3.1 Pro is a powerful, more reliable reasoning-and-coding engine you prompt for text and analysis. It is not a content studio — it drafts, reasons, codes, and reads images, but it renders no finished video, no branded image, and no scheduled post.
The clean way to think about K3.1 Pro is as a drafting-and-reasoning brain, and Kompozy as the studio and distribution floor that turns what it produces into finished, published content. This release sharpens the brain in two ways a creator can actually use: it got better at agentic, multi-step work, and more reliable (Moonshot reports lower hallucination), which makes it well suited to the research-heavy half of planning — paste an hour-long transcript, a research folder, or your last fifty posts into its 1M-token context and get back a vetted outline, a batch of angles, a script, or a newsletter draft. If you run it through Kimi Work or the API, it will even fan one source into a stack of drafts in a single agentic pass. What it hands you is still words and analysis: it renders no captioned vertical short, no face-locked avatar video, no brand-exact carousel, no quote card, and it posts nowhere.
That handoff is the entire job Kompozy does. Drop a K3.1 Pro draft or brief in as your source and Kompozy generates the finished media across 18 formats — Persona Shorts and HeyGen avatar video, Clipped Shorts, Carousels and Quote Graphics via brand-exact HyperFrames, Photo Posts, a Blog Article, and an Email Newsletter — every output held to one voice by your Persona Brief instead of reading like raw model output, then reframed to 9:16, 1:1, and 16:9 and scheduled across nine platforms plus blog and email from one queue with Autopilot. You don't even operate K3.1 Pro to do it: Kompozy runs its own managed Claude and OpenAI models for generation. So the practical division is this — use K3.1 Pro's long-context, agentic reasoning to think the idea through and get the research right, then let Kompozy produce the video, images, and posts and ship them. The model's speed and reliability improve the plan; Kompozy turns the plan into a published week.
Kimi K3.1 Pro is the first model in Moonshot AI's K3.1 series, launched in late October 2026. Moonshot describes it as a post-trained version of its Kimi K3 flagship — its most capable 3T-class open-weight model — with a 1-million-token context window and native multimodal input, served across kimi.ai, the Kimi apps, Kimi Work, Kimi Code, and the Kimi API. It is framed around coding, agentic work, 3D creation, and financial research.
Speed, not capability. Moonshot says Kimi K3.1 Pro Highspeed (API ID kimi-k3.1-pro-highspeed) runs roughly two to three times faster than K3.1 Pro with identical model capabilities, for more responsive interactions. Third-party listings put the Highspeed variant at roughly double the price and quota of the standard model.
Reported launch API pricing is near $2 per million input tokens and $10 per million output (cached input around $0.20), below K3's $3/$15 — but those figures come from third-party providers, so confirm on Moonshot's own pages. Moonshot framed it as an open-weight model with full weights slated to follow the hosted launch, pointing to on or around November 4, 2026.
No. K3.1 Pro drafts text and code, reasons over long inputs, and understands images, but it renders no captioned video, brand-consistent images, carousels, or scheduled posts. To turn a K3.1 Pro draft into finished, on-brand content, pair it with a content engine like Kompozy, which generates 18 formats and publishes to nine platforms plus blog and email.