Kimi-K2-Instruct: What PC You Need
Kimi-K2-Instruct needs about 699.2GB at Q4. No PC we track with a verified price runs it comfortably.
What Kimi-K2-Instruct asks of a PC
Kimi-K2-Instruct is a mixture-of-experts model with 1000B parameters, 32B of them active per token. Every expert has to stay resident, so its memory need follows the 1000B figure, not the 32B one.
At 8K context Kimi-K2-Instruct takes 699.2GB at Q4, 950.7GB at Q6 and 1223.9GB at Q8. Stepping from Q4 to Q8 costs 524.7GB.
Kimi-K2-Instruct accepts up to 128,000 tokens of context, and a long context adds memory: at 128K and Q4 it takes 729.2GB.
0 tracked PCs with a verified price run Kimi-K2-Instruct comfortably and 0 run it with limits.
Memory requirements for Kimi-K2-Instruct
| Quantization | 8K context | 32K context | 128K context |
|---|---|---|---|
| Q4 | 699.2GB | 705.2GB | 729.2GB |
| Q5 | 821.4GB | 827.4GB | 851.4GB |
| Q6 | 950.7GB | 956.7GB | 980.7GB |
| Q8 | 1223.9GB | 1229.9GB | 1253.9GB |
- Parameters
- 1000B
- Active per token
- 32B
- Context window
- 128,000
- Source
- Model card
Cheapest PCs that run Kimi-K2-Instruct comfortably
None we track with a verified price.
Cheapest PCs that run Kimi-K2-Instruct with limitations
None we track with a verified price.