Kimi K3.1: Features, Pricing and How to Use It

What is Kimi K3.1 Pro?

Kimi K3.1 Pro is a 3T-class AI model developed by Moonshot as the first entry in its K3.1 series. The model functions as a post-trained version of Kimi K3, trained using large-scale reinforcement learning across hundreds of downstream domains.

The model operates strictly with reasoning mode enabled, rejecting any requests where thinking is turned off. Moonshot offers both the standard K3.1 Pro model and Kimi K3.1 Pro Highspeed, which generates responses two to three times faster.

How does Kimi K3.1 Pro perform?

Kimi K3.1 Pro scores near leading proprietary frontier models across coding, spatial, and financial benchmarks. Artificial Analysis awarded K3.1 Pro an Intelligence Index score of 55, placing it above competitors like Gemini 4 Argon and GPT-6 Astra.

In internal evaluations, K3.1 Pro scored 61.2% on Kimi Code Bench 3.0 Long-horizon, improving on K3's 55.95%. Moonshot reported that claim-level hallucination rates dropped from 23.4% in K3 down to 8.3% on the Kimi Hallu Bench Hard stress test.

What are the core capabilities of Kimi K3.1 Pro?

Kimi K3.1 Pro handles multi-step software engineering, complex 3D creation, and multi-document financial workflows. The model writes CAD code, scripts parametric models for Blender, and integrates with game engines like Unity, Unreal Engine, and Godot.

• Software engineering: Solves multi-step terminal tasks, builds full web projects with Three.js, and runs long-horizon iterative simulation pipelines.
• 3D design: Creates printable assets, reconstructs 3D shapes from text or reference sketches, and powers the 3D Make tool on Kimi web and mobile apps.
• Financial research: Delivers end-to-end investment research, scored 64.4 on OfficeQA Pro v2, and analyzes data across spreadsheets.
• Cybersecurity controls: Moonshot applied training filtering to manage vulnerability research capabilities while maintaining standard programming utility.

How much does Kimi K3.1 Pro cost?

Kimi K3.1 Pro API pricing matches standard mid-tier frontier models. Standard API requests cost $2.00 per million input tokens and $10.00 per million output tokens.

For the Highspeed tier, cached input token hits remain priced the same as the regular model. Artificial Analysis measures the standard model's average generation speed at 73 tokens per second, noting it is more verbose than median models.

How to use Kimi K3.1 in official clients and APIs

Kimi K3.1 Pro is accessible via web interfaces, developer extensions, and OpenAI-compatible API endpoints. Users can select between regular and high-speed versions depending on workflow requirements.

1. Web and mobile apps: Access Kimi K3.1 Pro directly on Kimi.ai, Kimi Work, or through 3D Make in the Kimi mobile app.
2. Kimi Code Desktop: Open Composer and choose k3.1-pro or k3.1-pro-highspeed from the model picker. Restart the app if the models do not appear.
3. Kimi Code CLI: Switch active models by entering the /model command, or run /logout and /login if updates have not loaded.
4. VS Code extension: Pick the target model ID from the input bar dropdown menu.
5. Third-party tools and APIs: Generate an API key in the Kimi Code Console, set the model ID to k3.1-pro, and point to https://api.kimi.ai/coding/v1 or https://api.moonshot.ai/v1.

What are the limits of Kimi K3.1 Pro?

Kimi K3.1 Pro enforces strict operational constraints that developers must plan around. The model does not offer a non-thinking mode, meaning every generation incurs reasoning time and output overhead.

Switching models or changing reasoning effort levels within an existing chat session clears the context cache. Starting a new session is recommended whenever adjusting reasoning effort to avoid duplicate token billing.

Frequently asked questions

Sources: kimi.ai · Artificial Analysis · Simon Willison · Kimi Code Docs