Kimi K2 token cost & pricing
Kimi K2 is Moonshot's open-weight agentic model, priced at $0.60 per million input and $2.50 per million output with a 262K context window. It is tuned for tool use and multi-step task execution rather than pure chat.
Kimi K2 specifications
| Vendor | Moonshot |
|---|---|
| Input price | $0.6 / 1M tokens |
| Output price | $2.5 / 1M tokens |
| Context window | 262,144 tokens |
| Max output | 16,384 tokens |
| Open weights | Yes |
| Price rank | 8 of 17 (cheapest first) |
| Best for | agents, open-weights |
What Kimi K2 costs in practice
Four common workloads, priced at list rates. Monthly figures assume 30 days.
| Workload | Per request | Requests/day | Per month |
|---|---|---|---|
| Chatbot · 1K in / 500 out | $0.0019 | 10,000 | $555.00 |
| RAG Q&A · 8K in / 400 out | $0.0058 | 5,000 | $870.00 |
| Doc summary · 30K in / 1K out | $0.02 | 1,000 | $615.00 |
| Code agent · 60K in / 3K out | $0.04 | 500 | $652.50 |
How many requests does $100 buy?
Assuming a 1,000-token prompt and a 500-token response per request.
How Kimi K2 pricing actually works
A 4.2x input-to-output ratio at prices well below the Western frontier tier. Because the weights are open, hosted rates vary; the figures here reflect mainstream API pricing rather than a floor.
When to use Kimi K2
Worth testing for agent loops, function calling and tool-heavy automation where you want frontier-adjacent capability at a fraction of Claude Sonnet's price. Validate tool-call reliability on your own schemas — that is where agentic models differ most.
How to cut your Kimi K2 bill
Agent loops resend the full tool schema and history on every step, so cost grows superlinearly with step count. Cap iteration limits, prune old tool outputs from history, and cache the schema prefix where the host supports it.
Cheaper alternatives to Kimi K2
- Gemini 2.5 FlashGoogle10% cheaper
- DeepSeek R1 (reasoner)DeepSeek12% cheaper
- GPT-5 miniOpenAI27% cheaper
- Llama 4 MaverickMeta64% cheaper
Run the numbers on your own prompt
Paste your actual text into the token counter to get an exact count instead of an estimate. Nothing is uploaded.
Open the calculator →