Moonshot AI releases weights for Kimi K3.1 Pro model for independent deployment

Moonshot AI has launched Kimi Hosted Agents—a cloud service that packages the agent runtime environment, or 'harness,' as a managed product. This allows enterprises to avoid the need to build and maintain this infrastructure layer themselves, according to information from IT Home and the company's documentation on its open platform.

This offering is positioned as 'Harness as a Service.' Teams can configure agents on the Kimi infrastructure and receive a 24/7 environment for executing long-running tasks and collaborative work among multiple agents.

Enterprises can direct agents to internal systems for monitoring, research, and operational activities, or integrate them into customer-facing products as support or research assistants. Kimi Hosted Agents provides isolated sandboxes with network access controls, session boundaries that operators can stop at any time, and execution logs that can be exported for auditing. According to IT Home, the sandbox costs 0.072 yuan per hour plus model and plugin fees.

The platform currently includes over 140 skills and more than 30 plugins covering web search, professional data queries, office documents, and content generation. Developers can also connect their own tools, data sources, and MCP services. An additional security component, Kimi Inference Hook, allows the enterprise's proprietary or third-party security service to check task input data and tool output results before every model call, ensuring that only company-compliant content enters the inference process. This hook is currently available only to verified enterprise clients.

Official documentation also describes options for self-hosting environments, where sandbox tools run on customer-controlled machines while orchestration remains on the platform. In addition to the managed runtime, Moonshot is releasing the first series of official agent templates. IT Home notes eight financial templates covering modeling, deep research, morning briefings, IPO tracking, and portfolio monitoring; the documentation mentions a broader initial set of 12 templates, including game management and SRE incident response.

This launch expands Kimi's product line from models and desktop agents to a managed runtime, which Moonshot states the company will support and update. This occurred one day after Huawei Ascend announced support for zero-day inference for the open-weight Kimi K3.1 Pro model, highlighting a broader effort to simplify the deployment of Kimi agents in production environments.

Doubao Work, ByteDance's artificial intelligence agent application, announced on October 29th that it is expanding its functionality by providing access to third-party models for business clients. The first group of integrated models includes Moonshot AI's Kimi K3.1 Pro, Zhipu's GLM-5.3, and DeepSeek-V4.1-Flash. These models are now displayed in the application's model list alongside Doubao's proprietary models from ByteDance.

This update applies to team and enterprise subscription plans and is activated by default. Companies are not required to purchase separate access rights from each model developer or submit individual applications. Resource consumption is deducted from the AI credits that the organization already holds in Doubao Work.

Model Access Management

Administrators gain the ability to control which users can use a specific model. They can grant access to individual members, departments, groups, or the entire company. Furthermore, administrators can enable all models from one provider simultaneously or configure the use of a single specific model.

This move changes the positioning of Doubao Work. Previously, the application primarily served as a demonstration of ByteDance's Doubao models within an agent capable of interacting with files, browsers, and desktop applications. By allowing enterprises to direct tasks to competing Chinese models, the core product becomes the agent itself, rather than the underlying model. This also allows teams to choose a model based on the task type, such as using a coding-oriented model for engineering work and another model for handling long documents.

This development aligns with a broader trend among Chinese desktop agent products. This week, Xiaomi added DeepSeek, GLM, Kimi, and Qwen models to its MiMo Desktop, and several other agent tools have made similar moves. For enterprise-level buyers, a unified billing system and standardized permissions for multiple model developers significantly simplify the procurement process, which previously required comparing solutions from different manufacturers.

Doubao Work is continuously expanding its feature set. In recent weeks, capabilities such as saving recorded computer actions as reusable skills, importing chat history, skills, and memory from other AI tools with a single click, and an infinite creative canvas have been added. The application also supports parallel agents and local graphical interface control on Mac.

ByteDance has not specified what other third-party models will follow this initial batch, describing the current composition as the first group of models joining the platform.

Kimi, the chatbot from Moonshot AI, apologized to paying users for speed limitations and unjustified quota consumption when using the new K3.1 Pro model. The company decided to return part of each subscriber's monthly limit.

The company announced this early on October 26, according to sources IT Home and National Business Daily. In its statement, Kimi admitted that the problem was the company's fault and should have been resolved sooner.

As compensation, Kimi announced a one-time reset for all subscribers across all subscription tiers. Every active user will receive a refund equivalent to 25% of their total monthly quota, which is equivalent to about a week of usage. Users who used less than 25% of their quota will receive a reset of their usage to zero. Kimi specified that the adjusted values will be displayed in user usage indicators by the end of the day.

Furthermore, the company announced a configuration change that includes reducing unnecessary sub-agent calls and shortening output lengths. This will allow the same quota to perform more real tasks. The reason for the quota consumption issue was that in workflows simulating agent operation, one request could trigger multiple sub-agents and long responses, each consuming the user's limit.

These apologies came two days after Moonshot AI released K3.1 Pro on October 24 as the first model in the K3.1 series. The company positions it as its most powerful model to date, boasting around 3 trillion parameters and a context window of 1 million tokens. A high-speed version is also available, generating responses two to three times faster. The models are available on kimi.com, the Kimi app, Kimi Work, Kimi Code, and via the Kimi API. Moonshot AI plans to release the full model weights by November 4. The company reported a score of 65.7% on Terminal-Bench 4.0, a test for coding agents in real terminal environments, which is a self-reported figure.

Previously, Kimi had faced capacity overload. In July, shortly after launching K3, the company temporarily suspended new consumer subscriptions because demand exceeded available computing power.