Release tracker / Moonshot AI
Moonshot AI model releases
Every Moonshot AI model launch we have recorded, newest first: Kimi. Each one links to its source, and to how it ranks on our benchmarks where we have tested it.
- Latest
- Kimi K3, 16 Jul 2026
- Launches
- 9
- Since
- 20 Jan 2025
- Since last launch
- 88 days
Moonshot AI's latest model
-
#10 of 22 overall$3.00 in · $15.00 out per MReference source
Kimi K3 came out 88 days ago. Spotted a launch we're missing? Tell us.
How often Moonshot AI ships
Lines with at least three launches in the last two years. Amber: longer since the last launch than the line's average gap. Grey: no launch for 90 days or more, marked quiet.
| Line | Size | Launches | Average gap | Latest | Last launch |
|---|---|---|---|---|---|
| Kimi | Large | 9 | 68 days | Kimi K316 Jul 2026 | 88 days ago, longer than usual |
Every Moonshot AI release
-
#10 of 22 overall$3.00 in · $15.00 out per MReference source
-
Kimi K2.7 Code
-
$0.95 in · $4.00 out per MReference source
-
$0.57 in · $2.85 out per MReference source
-
$0.60 in · $2.50 out per MReference source
-
$0.60 in · $2.50 out per MReference source
-
$0.57 in · $2.30 out per MReference source
-
Kimi Dev
-
Kimi K1.5
Questions
- What is Moonshot AI's latest model?
- Kimi K3, released 16 Jul 2026 (88 days ago). It is the newest of the 9 Moonshot AI launches we have recorded since 20 Jan 2025.
- When will Moonshot AI release its next model?
- Moonshot AI has not announced its next model. It usually releases a new Kimi model about every 68 days; the last was Kimi K3, 88 days ago, longer than usual.
- How often does Moonshot AI release new models?
- Over the last two years: a new Kimi model about every 68 days (9 launches).
Other providers: OpenAI, Mistral AI, Alibaba, Anthropic, Google, DeepSeek, xAI, Meta, Z.ai, Microsoft, MiniMax, Xiaomi, Amazon, NVIDIA, StepFun. Or see every launch on one timeline.
A launch is the first day a model was publicly available, previews included; a family's sizes released together count once, and later dated updates of a model are left out. Main lines are checked by hand against the provider's own changelog or announcement where we could read it ("Provider"), otherwise a reference page ("Reference"); the rest come from Epoch AI's model data (CC BY 4.0).