Overview
Moonshot v1 is the first-generation API model line from Moonshot AI, the Beijing company behind the Kimi assistant, founded in March 2023 by Yang Zhilin, Zhou Xinyu and Wu Yuxin. It is the API counterpart to the Kimi chatbot that Moonshot AI launched in October 2023, and it became the company's standard developer offering before the open-weight Kimi K2 series arrived. The line is served on the Kimi / Moonshot platform through an OpenAI-compatible endpoint at api.moonshot.ai/v1.
The line ships as three text models that differ only by context length: moonshot-v1-8k, moonshot-v1-32k and moonshot-v1-128k, holding 8,192, 32,768 and 131,072 tokens respectively, with context (input plus output) being the only difference between them. Each has a matching multimodal sibling — moonshot-v1-8k-vision-preview, moonshot-v1-32k-vision-preview and moonshot-v1-128k-vision-preview — that reads image content and returns text. A routing alias, moonshot-v1-auto, automatically picks the cheapest context tier (8k / 32k / 128k) for each request based on its token count.
Moonshot AI has not publicly disclosed Moonshot v1's parameter count, training architecture, knowledge cutoff or formal benchmark scores, so this page documents the verifiable platform facts rather than estimates. The line is now a legacy generation: it remains listed and callable on the platform, but Moonshot AI steers new work toward its newer Kimi K2 models (K2.5 / K2.6 / K2.7-Code), which add native multimodality, larger 256K context and stronger agentic and coding capability.
| License | Proprietary |
|---|---|
| Weights | API only |
| Parameters | Not disclosed |
| Context | 8K / 32K / 128K (per variant) |
| Max output | Shared input+output budget within each variant's context (8,192 / 32,768 / 131,072 tokens) |
| Architecture | Proprietary decoder-only transformer (closed; design details not published) |
| Knowledge cutoff | Not officially published |
| Modalities | Text, Vision |
| Status | Legacy (still served on the Kimi/Moonshot platform; superseded by the Kimi K2 line) |
Pricing
| Input | $0.20 (8k) / $1.00 (32k) / $2.00 (128k) per 1M tokens (USD) |
|---|---|
| Output | $2.00 (8k) / $3.00 (32k) / $5.00 (128k) per 1M tokens (USD) |
Vision-preview variants are priced the same as the text model at each tier (8k/32k/128k). Prices exclude applicable taxes. moonshot-v1-auto bills at the rate of whichever tier it routes to.
Strengths
- Long-context coverage up to a 131,072-token window on moonshot-v1-128k, a headline feature of Moonshot AI from its 2023 launch
- Three context tiers (8K / 32K / 128K) so you can pay only for the window you need
- Vision-preview variants at every tier that accept image input and return text
- moonshot-v1-auto router that auto-selects the cheapest sufficient context tier per request
- OpenAI-compatible API (api.moonshot.ai/v1) that drops into existing OpenAI SDK code with minimal changes
Best for
- General-purpose Chinese and multilingual chat and text generation via API
- Long-document summarization, Q&A and analysis using the 128K-token variant
- Image understanding tasks (describe / extract text from images) with the vision-preview variants
- Cost-controlled deployments that route through moonshot-v1-auto to the smallest sufficient context tier
- Migrating an existing OpenAI-SDK application to a China-based provider with minimal code change
How to access
| Provider | Model ID |
|---|---|
| Moonshot AI (Kimi) Platform ↗ | moonshot-v1-8k / moonshot-v1-32k / moonshot-v1-128k (+ -vision-preview, moonshot-v1-auto) |
FAQ
What is Moonshot v1?
Moonshot v1 is Moonshot AI's first-generation API model line, the developer-facing counterpart to the Kimi assistant. It ships as the moonshot-v1-8k, moonshot-v1-32k and moonshot-v1-128k text models, plus matching vision-preview variants, served through an OpenAI-compatible API on the Kimi / Moonshot platform.
What is the difference between moonshot-v1-8k, 32k and 128k?
Per Moonshot AI's documentation, the three models differ only in maximum context length — 8,192, 32,768 and 131,072 tokens (input plus output) — with no difference in underlying model quality. You pick the tier that fits your prompt size, or use moonshot-v1-auto to route automatically.
Is Moonshot v1 open source, and how much does it cost?
No — Moonshot v1 is proprietary and closed-weight; only API access is offered. Published pricing (USD per 1M tokens) is $0.20 input / $2.00 output for the 8k tier, $1.00 / $3.00 for 32k, and $2.00 / $5.00 for 128k, with vision-preview variants priced the same as text at each tier.
Is Moonshot v1 still the model to use?
Moonshot v1 remains listed and callable on the platform, but it is now the older generation. Moonshot AI directs new projects to its newer Kimi K2 models (K2.5 / K2.6 / K2.7-Code), which add native multimodality, a 256K context window and stronger agentic and coding performance.