Overview
Kimi K3.1 Pro Highspeed is the faster-serving version of Kimi K3.1 Pro, a 3-trillion-parameter-class, open-weight model developed through additional post-training of Kimi K3. Moonshot AI says Highspeed retains the standard model’s capabilities while generating output approximately 2–3 times faster. It supports native visual understanding and a context window of 1,048,576 tokens. (kimi.ai)
Capabilities and uses
- Coding, debugging, refactoring, and long-running agent workflows across large codebases.
- 3D and spatial tasks, including generating scripts for CAD and other modeling tools.
- Financial research, document analysis, spreadsheet modeling, and multi-step work with tools.
- Interactive programming and other applications where users are waiting for generated output. (kimi.com)
Limitations
Outputs can contain factual or reasoning errors and should be checked, particularly for financial or other consequential decisions. Creating or modifying 3D assets and running code require appropriate tools and an execution environment. The stated 2–3× improvement concerns output generation and does not guarantee the same speedup for an entire task. Reasoning is always enabled; reasoning_effort supports low, high, and max, with max as the default. (kimi.com)
API pricing
The following are Moonshot AI’s pay-as-you-go list prices for kimi-k3.1-pro-highspeed, per 1 million tokens, before applicable taxes. Prices were checked on 2026-11-15. The domestic and international offerings are separate regional prices, not currency conversions of one package.
| Pricing category | Domestic API (CNY) | International API (USD) |
|---|---|---|
| Uncached input | ¥26.00 | $4.00 |
| Output | ¥136.00 | $20.00 |
| Cache hit / cached input | ¥1.30 | $0.20 |
| Cache write, 5-minute TTL | ¥26.00 | $4.00 |
| Cache write, 1-hour TTL | ¥52.00 | $8.00 |
The published rates apply across the context window without a context-length pricing tier. Cache hits, cache writes, and uncached input are separate billing categories; cache writes are not also charged as uncached input. The default cache TTL is five minutes. API usage is billed separately from product membership. (kimi.com)
Release and official sources
Release date: 2026-10-24. The official Chinese announcement is marked updated on that date and says Highspeed launched alongside K3.1 Pro. (kimi.com)