The cheapest paid Kimi K3 API access we can verify is $2.70 / $0.27 / $13.50 per million tokens (input, cached input, output): Wallaby's ongoing rate, 10% under official list on every column. Free routes exist too, and every one has a boundary. Here is where each one ends. As of October 2026.
Disclosure: Wallaby Token is our own service, so read this as the answer we can prove, not a neutral review. Rates below link to their sources.
The cheapest paid tier
Wallaby sells Kimi K3 at a flat 10% below Moonshot's official list ($3.00 / $0.30 / $15.00), on input, cached input and output alike. That is the standing rate, not a promo window, and it comes with per-request itemized billing and per-key spending caps. pricing.json carries the live numbers with an as-of date.
Several other providers, including Together, Fireworks and SiliconFlow, mirror the official list exactly. If you already live on one of them, you are paying list price for the same model.
Official direct, and when it wins
Moonshot's own platform charges list price and is where new checkpoints land first. If day-one access to new releases matters more than 10%, direct is the cleanest line. Two considerations: requests are processed in Beijing, which may not satisfy some enterprise data-residency requirements, and there is no free tier on the API.
Free options, and their boundaries
Three honest ones:
- Wallaby trial credit: $0.50 on every new account, no card, and it does not expire. Good for roughly 50 test calls. Its purpose is verification, not production.
- Kimi's consumer apps: the Kimi membership and Kimi Code are separate products with their own quotas. They are not API access and cannot serve your agent or pipeline.
- The official API: paid only. No free tier, full stop.
The catch columns
Headline price is not the whole price, and two columns decide what you actually pay.
The cache column. Coding agents resend most of their context every turn, so the cached-input rate carries the bill: in one session we metered, 93% of input tokens were cache reads. A provider that discounts input but holds cache at list can cost more than one at a flat 10% off. The blended spread between providers reaches 6× on the same workload, per the 17-provider comparison.
The billing shape. Per-request itemized lines, per-key caps and a public status page decide whether finance can audit your spend or just pay it. A mysteriously cheap endpoint with no itemization is a bill you cannot audit.
FAQ
What is the cheapest Kimi K3 API provider?
Among providers we can verify, Wallaby at $2.70 / $0.27 / $13.50 per million, ongoing. Aggregator listings show lower headline numbers from time to time; before trusting one, check whether the low price applies only to uncached input, whether cached tokens or cache writes stay at full list, and whether the route supports context caching at all — a cheap endpoint with no caching can cost an agent workload more than list price does.
Is there a free Kimi K3 API?
No official free tier exists. The closest is trial credit: $0.50 on a new Wallaby account, billed per request from the first call.
Is Kimi K3 the cheapest coding model?
No, and it does not try to be. Budget tiers like DeepSeek V4.1-Flash run at $0.30 / $1.20 per million at peak. K3 sits at the reasoning tier; the model-level comparison lays out what the extra money buys.
Do discounted providers keep the cache discount?
Ours does: cached input is $0.27, the same 10% off. Elsewhere, check row by row. The cache column is where discounts quietly disappear.
Sources
- pricing.json: Wallaby rates, live
- Kimi K3 access routes compared: provider-by-provider notes, first-hand
- platform.kimi.ai: official list
Where to go next
You came for the cheapest route, so the short version is: $2.70 / $0.27 / $13.50 at Wallaby, or free via the $0.50 trial credit if you are still evaluating. Three next steps:
- Wider: Kimi K3 API pricing across 17 providers — the full table.
- Deeper: what a month actually costs — our own metered bill.
- Start: free trial credit, no card — first call in two minutes.