Markdown edition

# Kimi K3.1 Pro reaches hosted services before full weights release

> Kimi K3.1 Pro and Highspeed are available through Kimi products, the API and Vercel AI Gateway. Full weights are promised by November 4; access conditions, prices and vendor test limits matter now.

By BIGCHANGE.AI Editorial

Published: 2026-10-24T12:38:39.475Z
Updated: 2026-10-24T12:38:39.475Z
Canonical: https://bigchange.ai/blog/kimi-k31-pro-hosted-access-weights-release

![Conceptual charcoal illustration of an orange network cable plugged into a rack appliance beside an open, empty drive bay.](https://bigchange.ai/api/media/file/kimi-hosted-access-empty-local-bay-hero-v2.png)
AI-generated conceptual illustration by BIG CHANGE; no actual Kimi hardware, interface, deployment, download or test is depicted.

Kimi has released **K3.1 Pro and K3.1 Pro Highspeed** across its hosted products and API. Developers can also select both through Vercel AI Gateway, whose [October 23 announcement](https://vercel.com/changelog/kimi-k3-1-pro-and-kimi-k3-1-pro-highspeed-are-available-now-in-ai-gateway) provides the clearest dated record of availability. Kimi describes K3.1 Pro as an open-weight model, but says the **full weights will be released by November 4, 2026**. For now, the confirmed access is hosted.

## The big change

- **What changed:** Kimi made K3.1 Pro and a faster Highspeed variant available through Kimi.ai, its apps, Kimi Work, Kimi Code and its API. Vercel added both to AI Gateway on October 23.
- **Why it matters:** Teams can evaluate the new model through existing hosted interfaces and compare the variants' speed and token prices. Downloading the complete weights and running them on their own infrastructure remains a promised later step.
- **What to watch:** Kimi has committed to releasing full weights by November 4. Its launch material does not establish independent performance or reliability in a reader's own workload, and Kimi Enterprise administrators must enable access for their organizations.

## Hosted access now, weights later

Kimi's [release post](https://www.kimi.ai/blog/kimi-k3-1-pro) says both variants are available on Kimi.ai, Kimi Work, Kimi Code and the Kimi API. The availability section also names Kimi mobile apps. Kimi Work requires desktop version **3.1.0 or later** on Windows or Apple silicon Macs. In Kimi Code, users select the model with `/model`. Kimi Enterprise access is **off by default**; an administrator must enable it. Those conditions matter more immediately to a prospective user than the eventual weights release.

For direct API calls, Kimi's [quickstart](https://platform.kimi.ai/docs/get-your-api-key) specifies an API key and the model IDs `kimi-k3.1-pro` and `kimi-k3.1-pro-highspeed`. It documents Chat Completions, Responses and Anthropic-compatible Messages interfaces, including a playground. [Vercel AI Gateway](https://vercel.com/changelog/kimi-k3-1-pro-and-kimi-k3-1-pro-highspeed-are-available-now-in-ai-gateway) uses its own prefixed IDs: `moonshotai/kimi-k3.1-pro` and `moonshotai/kimi-k3.1-pro-highspeed`. Vercel lists AI SDK, Chat Completions, Responses, Anthropic Messages and connected coding agents as routes. The two ID schemes should not be interchanged.

Kimi's API platform [currently lists](https://platform.kimi.ai/) K3.1 Pro at **$2 per million input tokens and $10 per million output tokens**. Highspeed is **$4 input and $20 output**. Both list a **$0.20 cache-hit price** per million tokens; the displayed cache-write prices are $2 and $4 respectively. Kimi's [pricing explanation](https://platform.kimi.ai/docs/pricing/chat) says K3-series caching is automatic for repeated prefixes, with a default five-minute lifetime, and that quoted prices exclude applicable taxes. These are API prices, not Kimi app or Work subscriptions. Kimi says Highspeed delivers roughly **two to three times** K3.1 Pro's speed with the same capabilities; BIG CHANGE has not measured latency or throughput.

Both variants offer a **1,048,576-token context window** according to Kimi's pricing documentation. That is a capacity specification, not evidence that a million-token request will be accurate, affordable or fast for a particular task. The exact cost depends on input, output and cache use.

## What the capability evidence shows

K3.1 Pro is a post-trained version of K3, according to Kimi. Its release post emphasizes coding, 3D creation and financial research. One useful comparison is Kimi's **61.2%** score on its internal Kimi Code Bench 3.0 Long-horizon, against **55.95%** for K3. That is evidence of improvement under Kimi's chosen evaluation, not an independent result. The post's cross-model scores combine different harnesses and sources; its footnotes say K3.1 Pro commonly uses maximum reasoning effort, and some other model scores come from official leaderboards or vendors' own reports. A buyer should test the hosted model on the tasks and settings that matter to them.

Kimi also reports a targeted cyber capability intervention. It says it removed cyber-related material from mid-training data and filtered cyber-related reinforcement-learning tasks for the planned open-weight release. On its internal vulnerability benchmark, the resulting K3.1 Pro checkpoint solved **81.0%** of problems at pass@3, compared with **90.5%** for a checkpoint without those controls. The company still reports more cyber capability than K3. These figures describe Kimi's benchmark and training comparison; they do not prove that misuse is prevented. Kimi's stated policy is to avoid releasing open weights that extend the absolute frontier of offensive cyber capability.

The staged release is the practical boundary. Hosted use, published prices and model IDs are available to assess today. Full-weight availability and self-hosting await the promised release, while real-world capability, speed and safety need evaluation outside the vendor's reported tests.

## Sources & further reading

- [Kimi: Introducing Kimi K3.1 Pro](https://www.kimi.ai/blog/kimi-k3-1-pro): provider account of launch channels, Work and Enterprise prerequisites, November 4 weights promise, benchmark methods and cyber training controls. The retrieved page does not display a publication date.
- [Kimi API quickstart](https://platform.kimi.ai/docs/get-your-api-key): current model IDs, API key prerequisite, compatible interfaces and playground.
- [Kimi API platform](https://platform.kimi.ai/) and [pricing explanation](https://platform.kimi.ai/docs/pricing/chat): listed token and cache prices, context size, billing basis and tax qualification. Prices can change.
- [Vercel: Kimi K3.1 Pro and Highspeed in AI Gateway](https://vercel.com/changelog/kimi-k3-1-pro-and-kimi-k3-1-pro-highspeed-are-available-now-in-ai-gateway): dated October 23, 2026; gateway-specific IDs and supported routes.
- [BIG CHANGE: vLLM's Kimi K3 decode-path benchmark](https://bigchange.ai/blog/kimi-k3-vllm-gpu-inference-benchmark): earlier, distinct reporting on a K3 serving test, rather than this K3.1 hosted launch.

## Sources

- [Kimi, Introducing Kimi K3.1 Pro](https://www.kimi.ai/blog/kimi-k3-1-pro) — Provider release stage, channels, prerequisites, weight date, benchmarks and cyber controls; retrieved page undated.
- [Kimi API Platform](https://platform.kimi.ai/) — Current displayed API token and cache prices.
- [Kimi API quickstart](https://platform.kimi.ai/docs/get-your-api-key) — Direct API model IDs, key prerequisite and interfaces.
- [Kimi API pricing explanation](https://platform.kimi.ai/docs/pricing/chat) — Context, caching, pay-as-you-go billing and tax qualification.
- [Vercel, Kimi K3.1 Pro and Highspeed in AI Gateway, October 23, 2026](https://vercel.com/changelog/kimi-k3-1-pro-and-kimi-k3-1-pro-highspeed-are-available-now-in-ai-gateway) — Dated Gateway availability, IDs and interfaces.
- [BIG CHANGE, vLLM Kimi K3 benchmark](https://bigchange.ai/blog/kimi-k3-vllm-gpu-inference-benchmark) — Related but distinct earlier coverage.