Moonshot AI
Kimi K2.6 (1T)
Open-weight MoE competing with frontier closed models
Model Summary
Family
Kimi
Version
2.6
Parameters
1000B (est.)
Parameter counts for closed models are estimates; vendors rarely publish exact sizes.
VRAM Requirements by Quantization
Memory needed to serve Kimi K2.6 (1T) for inference, including a 20% overhead for activations and KV cache.
| Precision | VRAM needed | Smallest single GPU that fits |
|---|---|---|
| INT4 (4-bit) | 558.79 GB | Multi-GPU required |
| INT8 (8-bit) | 1117.59 GB | Multi-GPU required |
| FP16 (16-bit) | 2235.17 GB | Multi-GPU required |
| FP32 (32-bit) | 4470.35 GB | Multi-GPU required |
Recommended GPU Configurations
Cheapest on-demand configurations to serve Kimi K2.6 (1T) at 8-bit (1118 GB VRAM).
9x AMD Instinct MI250X
1152 GB total VRAM · CDNA 2 · multi-node
~$22.50/h
70x NVIDIA T4
1120 GB total VRAM · Turing · multi-node
~$35.00/h
6x AMD Instinct MI300X
1152 GB total VRAM · CDNA 3
~$36.00/h
Quick GPU Planning
Use the calculator pre-filled with this exact version to estimate memory, speed, and compute requirements in a few clicks.
Access Pre-filled CalculatorFrequently Asked Questions
How much VRAM do you need to run Kimi K2.6 (1T)?
With an estimated 1000B parameters, Kimi K2.6 (1T) needs roughly 1118 GB of VRAM in 8-bit (INT8), 559 GB in 4-bit, and 2235 GB in FP16, including a 20% overhead for activations and KV cache.
Which GPUs can run Kimi K2.6 (1T)?
At 8-bit quantization, the most cost-effective option is 9x AMD Instinct MI250X (1152 GB combined VRAM, around $22.50/hour on-demand). Higher-end cards like the NVIDIA B200 or AMD MI355X reduce the GPU count needed.
Can Kimi K2.6 (1T) run on a single GPU?
No. Even in 4-bit, Kimi K2.6 (1T) exceeds the memory of any single current GPU, so a multi-GPU cluster is required.
How much does it cost to serve Kimi K2.6 (1T) in the cloud?
Renting 9x Instinct MI250X costs on the order of $22.50/hour, i.e. about $16,425/month running 24/7. Actual prices vary by provider and commitment; spot and reserved capacity can be significantly cheaper.
Deploy on a GPU cloud
Rent 9x Instinct MI250X by the hour instead of buying hardware.
