Kimi K3
Moonshot's open flagship. Index 44 at max, output around 34 tokens per second, which is the slow side of this board.
Open weights
44Index · Max
$2.00Per measured task
22.0 pts/$Points / $
34Tokens / sec
No High score is published. Showing Max.
Every published effort
| Setting | Index | Per task | Tokens / sec |
|---|---|---|---|
| Max | 44 | $2.00 | 34 |
| Low | 30 | $1.15 | 36 |
What this workload costs
List price for one call. Reasoning tokens count as output. A cache hit uses the lab’s cache rate when one is published.
Kimi K3
No cache priceSpecs
- API id
- Not published here
- Released
- Sep 2026
- Context
- 1.05M
- Max output
- —
- Input / 1M
- $3.00
- Output / 1M
- $15.00
- Cache read / 1M
- —
- Knowledge
- —
- Inputs
- Text
- Weights
- Kimi K3 license, not MIT. Read it before you embed the weights.
Vendor list price as compiled on 23 Sep 2026. A yuan conversion can print a few cents lower. About 2.8T total, 104B active. On the index by 7 Sep 2026. The day is not more precise here.
Use it when
An open flagship with a 1.05M window, when the Kimi license fits and you want a score within a few points of the closed frontier.
Leave it when
A tight API budget. At $3 / $15 it is priced like a closed model. MiMo-V2.6-Pro scores higher here and costs less per task.