Every Kimi AI model, compared
Moonshot AI ships five callable Kimi models as of August 2026. K3 is the flagship with a 1,048,576-token context window; K2.7 Code is the cheapest good option for coding. Two older models are being retired. Every figure below is taken from Moonshot's own documentation.
Last updated
At a glance
Specifications and pricing
USD per million tokens, excluding tax. Cache-hit pricing applies automatically to repeated prompt prefixes.
| Model | Context | Input (hit) | Input (miss) | Output | Status |
|---|---|---|---|---|---|
| Kimi K4 | — | — | — | — | upcoming |
| Kimi K3 | 1,048,576 | $0.30 | $3.00 | $15.00 | current |
| Kimi K2.7 Code | 262,144 | $0.19 | $0.95 | $4.00 | current |
| Kimi K2.7 Code HighSpeed | 262,144 | $0.38 | $1.90 | $8.00 | current |
| Kimi K2.6 | 262,144 | $0.16 | $0.95 | $4.00 | current |
| Kimi K2.5 | 262,144 | $0.10 | $0.60 | $3.00 | legacy |
| Kimi K2 | — | — | — | — | deprecated |
Models
Available now
Open to new API users with published pricing.
Kimi K3
1MKimi K3 is Moonshot AI’s flagship model, with 2.8 trillion parameters, native visual understanding and a 1,048,576-token context window.
kimi-k3
Kimi K2.7 Code
256KKimi K2.7 Code is Moonshot AI’s coding-focused model, with a 262,144-token context window and higher instruction-following reliability in long contexts.
kimi-k2.7-code
Kimi K2.7 Code HighSpeed is the same model as Kimi K2.7 Code served at roughly 180 tokens per second, at double the price.
kimi-k2.7-code-highspeed
Kimi K2.6
256KKimi K2.6 is a general-purpose model with a 262,144-token context window, supporting text, image and video input across thinking and non-thinking modes.
kimi-k2.6
Models
Closing soon
Still callable, but closed to new accounts or scheduled for shutdown.
Kimi K2.5
256KKimi K2.5 is a multimodal model with a 262,144-token context window. It is closed to new API users and the platform sunsets it on August 31, 2026.
kimi-k2.5
Models
Discontinued
The hosted endpoints no longer respond.
Kimi K2 was Moonshot AI’s open-weights model family from 2025. All K2 API variants were discontinued on May 25, 2026.
Models
Unannounced
No announcement from Moonshot AI. Nothing here is confirmed.
Kimi K4 is an unannounced successor to Kimi K3. Moonshot AI has not confirmed its existence, specifications, pricing or release date.
FAQ
Questions about the Kimi lineup
How many Kimi models are there?
Moonshot AI currently offers five callable Kimi models: K3, K2.7 Code, K2.7 Code HighSpeed, K2.6 and K2.5. K2.5 is closed to new accounts and sunsets on August 31, 2026. The older kimi-k2 series was discontinued on May 25, 2026.
Which Kimi model should I use?
Use Kimi K3 for the hardest reasoning and long-context work — it is the only model with a 1M-token window. Use Kimi K2.7 Code for day-to-day coding at roughly a quarter of K3’s output price. Use Kimi K2.6 for general-purpose work where you do not need the coding specialisation.
What is the largest Kimi context window?
Kimi K3 has the largest context window at 1,048,576 tokens. Every other current Kimi model — K2.7 Code, K2.7 Code HighSpeed, K2.6 and K2.5 — has a 262,144-token window.
Do Kimi models support images and video?
Kimi K3 has native visual understanding. Kimi K2.7 Code, K2.6 and K2.5 accept text, image and video input. The exact modality support per model is listed on each model page.
What happens when a Kimi model is discontinued?
The hosted API endpoint stops responding. Open weights, where Moonshot published them, remain downloadable — so a discontinued model can still be self-hosted. That is the case for the kimi-k2 series, discontinued on May 25, 2026.