The Kimi K3 family
K3 is the only released member of this family, and its biggest change over the previous generation is the context window: from 256K tokens on K2.6 and K2.7 Code to 1,048,576. If you feed long text, that single number settles the choice.
- one member, a one row timeline
- four times the previous window
- 2.8 trillion parameters
Last checked: This is a reference page. It is re-checked against the vendor sources and updated when a new version ships. What changed
Which catalogue model to pick
The Moonshot catalogue holds three models and choosing between them comes down to one thing: input length. K3 carries a 1,048,576 token window; K2.7 Code and K2.6 both stop at 256K. If the job is a whole repository or several long documents, K3 is the only option. If your input is short, K2.7 Code, presented for programming and offered in a high-speed variant, looks like the sensible pick.
There is a large qualification. Moonshot publishes no price at all for K2.7 Code or K2.6. The one pricing page on the platform covers K3. So the usual advice, take the smaller model for the smaller job, cannot be backed with a number here, and we do not write advice we cannot cost.
The timeline of this family is one row
K3 was announced on the Moonshot site on 16 July 2026, and no second member has joined the line since. A family page normally exists so you can see what changed between versions. Here nothing has changed yet, and saying so beats manufacturing a timeline.
One thing was never a member of this family: K3 Max. That is a leaderboard row, and no model of that name sits in the Moonshot catalogue.
A price that does not line up with the window
K3 wins the context column outright in our coding table, because it holds the largest number there. Yet its lead over the one million tokens of Claude and GLM is under five percent, and that same under five percent hands it the whole weight of the column. This is a weakness in our normalisation, and we would rather write it down than have somebody find it.
Release timeline
- Kimi K3 Max current The cheapest of the top three in our coding table, and the largest context window in it.
Using it from Iran
This column carries a date and says how it was checked, because most listicles guess it. Where we have only read a vendor policy page, the note below says exactly that.
- Reachable
- not checked
- Payment
- not checked
- Free tier
- no
How we checked: Moonshot publishes no supported-countries list, so we can write neither open nor blocked. We have not tested from an Iranian connection either.
What it is good at
- The largest context window in the Moonshot catalogue and in our whole coding table, at 1,048,576 tokens
- $3 in and $15 out, roughly forty percent below Opus 5
- A cached input rate of $0.30 for workloads with a fixed prefix
- Native visual understanding, as described on the platform itself
Where it falls short
- The family has one member, so there is no version to version comparison to make.
- No price is published for the other two catalogue models, so nobody can say how much the smaller model saves.
- No model called K3 Max exists in the catalogue; that name lives only on the leaderboard.
- Access from Iran is unchecked, and Moonshot publishes no country list.
Our take
If you are building on Moonshot, K3 is effectively the only serious choice, and pushing long input at it costs nothing extra. For short work, until a price for K2.7 Code appears, any saving is a guess rather than a calculation.
Questions people actually ask
Is K3 better than K2.7 Code for programming
Moonshot documentation presents K2.7 Code for programming and K3 for long-horizon coding and deep reasoning. It publishes no comparative figure between them, so the only measurable difference is the context window: 1,048,576 against 256K tokens.
Sources
- Kimi platform models vendor source 12 August 2026
- Kimi K3 pricing vendor source 12 August 2026
- Moonshot AI vendor source 12 August 2026