A price per million tokens is hard to picture. This page turns Kimi's API prices into real costs: what one chat reply, one document or 1,000 requests cost, what $10 buys, and the hidden things that make bills bigger than expected. Enter your own numbers in the estimator to plan a budget.
An illustration: the cost of one short reply, worked out at list prices.
A short reply here means about 1,000 tokens in and 300 out. Prices from Moonshot's pricing pages, read 8 October 2026.
Three request sizes, on each model, with no caching.
| Model | Short reply 1,000 in · 300 out | Long answer 5,000 in · 1,000 out | Big document 50,000 in · 2,000 out |
|---|---|---|---|
| kimi-k3 | $7.50 | $30.00 | $180.00 |
| kimi-k2.7-code | $2.15 | $8.75 | $55.50 |
| kimi-k2.7-code-highspeed | $4.30 | $17.50 | $111.00 |
| kimi-k2.6 | $2.15 | $8.75 | $55.50 |
On a small screen, swipe the table sideways to see every column.
Standard list prices before tax. Kimi K3 and K2.7 Code think before answering, and that thinking counts as output, so their real output is often larger.
Pick a job to fill in typical numbers, then change anything to match yours.
| Same job on | Per request | A month |
|---|
Web searches are priced as Web Search Basic, $0.002 each. Batch is listed only for Kimi K2.6 and K2.7 Code. Kimi K3's one-off charge for saving text to the cache is left out. The job presets are our own rough assumptions.
How much does it cost to send a text to Kimi, or have Kimi write it?
| Model | Kimi reads it | Kimi writes it |
|---|
Rough rule: 1,000 English words ≈ 1,300 tokens, based on Moonshot's guide that a token is 3 to 4 characters. A novel of 90,000 words is about 117,000 tokens.
The API does not remember earlier messages. To keep a conversation going, your app sends the whole history again every time. So message 20 costs far more than message 1.
Assumes each user message is 150 tokens and each answer 400 tokens. "With caching" charges the repeated history at the cached rate, which Moonshot applies automatically when requests start with the same text; real cache hits can be lower.
| Trick | Saving | Example |
|---|---|---|
| Use Kimi K2.6 instead of K3 | About 70% | 1,000 short replies: $7.50 down to $2.15 |
| Cache repeated input | 80% to 90% off that input | Keep fixed instructions first and identical |
| Batch work that can wait | 40% | K2.6 or K2.7 Code only |
| Cap output length | Varies | Set a maximum on every request |
| Lower thinking effort | Varies | Low effort on K3, or thinking off on K2.6, for simple jobs |
| Trim chat history | Varies | Send a summary instead of every old message |
On a small screen, swipe the table sideways to see every column.
Percentages from Moonshot's list prices. More detail: Kimi API pricing · API or membership?