Kimi K3.1 Pro bills cached input at $0.20 per million tokens. A miss costs $2.00, ten times more. Any workload that resends the same prefix (an agent's system prompt, a long repository map, a document you keep asking about) lives on that ratio, not on the headline input rate.
Disclosure: Wallaby Token sells K3.1 Pro at $1.80 input, $9.00 output and $0.18 cached input, 10% under Moonshot's list. The official numbers come from Moonshot's pricing page; the arithmetic below is ours.
The rate card
| Per 1M tokens | Official | Wallaby |
|---|---|---|
| Input (cache miss) | $2.00 | $1.80 |
| Output | $10.00 | $9.00 |
| Cached input (cache hit) | $0.20 | $0.18 |
Every hit saves $1.80 per million tokens against the miss price on official rates, and $1.62 on ours. Moonshot bills cache writes separately by TTL (5 minutes by default, 1 hour on request), and only the hit is charged at the cached rate; the context-caching guide has the TTL mechanics.
What a hit rate is worth
Take an agent turn that resends 50,000 input tokens and gets 4,000 back. At official rates, with no cache, input costs $0.1000. At a 90% hit rate the same 50,000 tokens cost $0.0100 in misses plus $0.0090 in hits — $0.0190, an 81% cut on the input side.
| Hit rate | Input cost, 50K tokens | Blended per 1M | Saved vs no cache |
|---|---|---|---|
| 0% | $0.1000 | $2.00 | 0% |
| 50% | $0.0550 | $1.10 | 45% |
| 90% | $0.0190 | $0.38 | 81% |
| 93% | $0.0146 | $0.292 | 85% |
| 99% | $0.0038 | $0.218 | 89% |
The blended price at 93% is $0.292 per million, close to the $0.20 floor, which is why the last few points of hit rate are worth less than the first fifty. If your agent already sits above 90%, chasing 99% is optimization theatre; the win is in getting from 0% to 50%.
When caching will not help
Three cases, all common:
- One-shot requests. A single question against a fresh document writes the cache and never reads it. You pay the write, not the hit.
- Prefixes that change every turn. Cache matching is prefix-based. Put the volatile part (the new user message, the fresh tool output) at the end, or the prefix never matches.
- Gaps longer than the TTL. The default 5-minute window covers a working session; an agent that idles for ten minutes starts from a miss. A 1-hour TTL costs more per write and pays only if the reuse actually happens inside the hour.
Available now
- Model IDs:
kimi-k3.1-proandkimi-k3.1-pro-highspeed, served athttps://api.wallabytoken.com/v1behind an OpenAI-compatible chat-completions endpoint. - Context: 1,048,576 tokens, with reasoning output billed as output.
- Cache: repeated prefixes bill at $0.18 per million on both variants.
- Price: $1.80 / $9.00 per million input / output tokens on the standard variant, $3.60 / $18.00 on Highspeed — a flat 10% below Moonshot's list, standing rate with no promo code and no end date.
- Start: $0.50 trial credit on sign-up with no card, then top-ups by card, Apple Pay or Google Pay.
The number to watch
Divide cached input tokens by total input tokens per request, then look at the trend across a session. Under 40%, fix the prompt layout before touching anything else. Over 90%, stop optimizing the cache and look at output: on K3.1 Pro, output is five times the input rate, and a chat turn with a 90% hit rate still spends most of its money on the 500 tokens it writes back. The worked example is in the pricing guide; the K3 version of this math, with Moonshot's write fees, is in the K3 cache post.