Kimi K3.1 Pro lists at $2.00 per million input tokens, $10.00 per million output tokens and $0.20 per million cached input tokens. Wallaby's price when the endpoint opens is $1.80, $9.00 and $0.18, 10% below that list.
| Per 1M tokens | K3 official | K3.1 Pro official | Wallaby K3.1 Pro |
|---|---|---|---|
| Input | $3.00 | $2.00 | $1.80 |
| Output | $15.00 | $10.00 | $9.00 |
| Cached input | $0.30 | $0.20 | $0.18 |
Output costs five times input per token. That ratio, not the input rate, decides most agent bills.
Three worked examples
All three use list prices and round to four decimals. They exclude taxes and retries.
| Workload | Tokens | K3 official | K3.1 Pro official | Wallaby K3.1 Pro |
|---|---|---|---|---|
| Chat turn, 90% cache hit | 8,000 in, 500 out | $0.0121 | $0.0080 | $0.0072 |
| Agent turn, 70% cache hit | 50,000 in, 4,000 out | $0.1155 | $0.0770 | $0.0693 |
| Document pass, no cache | 1,000,000 in, 100,000 out | $4.50 | $3.00 | $2.70 |
The document pass is the easiest to check. One million input tokens at $1.80 is $1.80. One hundred thousand output tokens at $9.00 is $0.90. Total, $2.70.
The chat turn is the one that surprises people. Even with a 90% cache hit rate, the 500 output tokens cost $0.0045 of the $0.0072 total. Output is 62% of that bill.
Where caching changes the answer
Cache hits are billed at the cached input rate. On an 8,000-token prompt with 90% hits, you pay for 800 fresh tokens at the input rate and 7,200 at the cached rate:
800 x $1.80 / 1M = $0.00144
7,200 x $0.18 / 1M = $0.00130
500 x $9.00 / 1M = $0.00450
Total = $0.00724
That is a 40% cut against the same turn at K3's official rates. The saving comes from both the lower input rate and the lower output rate; neither line alone explains it.
What a month looks like
At 10,000 agent turns a month, the 50,000-in / 4,000-out example costs about $693 at our rate card and $1,155 at K3's official rates. At 1,000 turns, it is $69 and $116.
These are list-price estimates on a fixed workload. Your mix of input, output and cache hits will move the number. The invoice is the authority.
Two pricing details worth knowing
Moonshot prices batch inference at 60% of the standard rate for kimi-k3.1-pro and kimi-k3. On the document pass above, batch takes the official cost from $3.00 to $1.80.
Moonshot also sells a highspeed variant of K3.1 Pro at 2-3x the speed of the standard model. Official price is $4.00 input, $20.00 output and $0.20 cached input. Our retail price for that variant is not set yet, so the table above covers the standard model only.
The short version
Input rates are the headline. Output rates are the bill. At $1.80 input and $9.00 output, K3.1 Pro is 40% below K3's official rates on both lines, and the cached rate drops from $0.30 to $0.18.
Our endpoint opens shortly. The release write-up covers what changed in the model, and the 5,315-task test covers how it scored.