What Moonshot AI actually builds
Moonshot AI is the Chinese company behind Kimi, and its platform lists exactly three models today: K3, K2.7 Code and K2.6. Only K3 has a pricing page, at $3 in and $15 out per million tokens, and it is the model sitting third in our coding table.
- three models in the whole catalogue
- 1,048,576 token window on K3
- no supported-countries list
Last checked: This is a reference page. It is re-checked against the vendor sources and updated when a new version ships. What changed
Model families
One model line with a page. The other two catalogue entries stayed on the previous generation.
Versions
| Versions | Released | Status | Model families |
|---|---|---|---|
| Kimi K3 Max | current | Kimi K3 |
The catalogue really is three models, and that says something
The Kimi platform model page lists three things and stops: K3 with a one million token window and 2.8 trillion parameters, K2.7 Code at 256K aimed at programming, and the general purpose K2.6, also 256K. Google publishes dozens of models. Alibaba lists separate embedding and reranking models. Moonshot builds one frontier model and spends the rest of its effort on the product around it.
The practical effect is that choosing is easy. Long documents leave you one option, because the other two carry a quarter of the window.
A name that is on the leaderboard and not in the catalogue
The Arena row for this maker is called kimi-k3-max. No such model exists in Moonshot documentation: the API id is plain kimi-k3, and the max suffix names a reasoning effort level. We repeat it here because most Persian round-ups have presented Kimi K3 Max as a separate product with a separate price, and there is nothing of that name to buy.
A second rate that only pays off on a repeated prompt
Moonshot publishes two input prices: $3 on a cache miss and $0.30 on a hit. That is a tenfold gap. The lower number reaches you only when a long fixed prefix repeats on every request, such as a system guide of several thousand tokens. If your workload sends fresh text each time, the real number is $3, and budgeting on the cache rate multiplies the invoice that actually arrives.
What it is good at
- The cheapest route to a top-tier coding model: $3 and $15 against $5 and $25 for Opus 5, eighteen points behind it on the WebDev leaderboard
- The largest context window among the four Chinese models in our table, at 1,048,576 tokens
- An OpenAI-compatible interface, so code written for that stack runs unchanged
- A cached input rate of $0.30 for workloads built on a fixed prefix
Where it falls short
- It publishes no supported-countries list, so our Iran column stays unchecked and earns nothing in the ranking.
- The platform documentation carries no official command line tool and no editor extension. Its tooling maturity scores 1 of 4.
- No price is published for K2.7 Code or K2.6, so you cannot even make a cost choice inside its own catalogue.
- There is no image, video or music model anywhere in the platform documentation.
- No model called Kimi K3 Max exists in the catalogue, whatever the leaderboard row and the round-ups say.
Our take
If you write code and the invoice matters, Moonshot is the most serious cheaper alternative to Anthropic, and its context window is the biggest of the group. If you need official tooling and a support contract, this is still an API and nothing more.
Questions people actually ask
Is Kimi K3 Max different from Kimi K3
The Moonshot catalogue holds only kimi-k3. What the leaderboard calls kimi-k3-max is the same model at a higher reasoning effort, not a separate product, and the price and context window belong to the base model.
Does Kimi work from Iran
We do not know, and we have not written it down. Moonshot publishes no supported-countries list and we have not tested from an Iranian connection yet. Any list that answers this confidently has no source for it.
Sources
- Kimi platform models vendor source 12 August 2026
- Kimi K3 pricing vendor source 12 August 2026
- Moonshot AI vendor source 12 August 2026