Kimi K3
A large-scale vision-language MoE for coding and long-horizon agent tasks.
- Published size
- 2.8T
- Active parameters
- 104B
- Context window
- 1M
- Architecture
- MoE · vision
- License
- Kimi K3 License
Deployment considerations
Official quantization and distributed serving requirements are central to deployment planning at this scale.
The model summary lists 2.8T total and 104B activated parameters, MXFP4 weights, MXFP8 activations, and a 1,048,576-token context.
GPU memory estimates and minimum / recommended configurations are pending. This profile does not contain measured deployment results.
GPU requirements
- Minimum
- —GPU / VRAM · pending
- Recommended
- —GPU / VRAM · pending
Memory estimates and tested configurations will appear here after checkpoint review and deployment testing.
No invented leaderboards.
Latency, throughput, and cost per token will appear here after a reproducible run. Until then, this page helps you understand the model—not predict its performance.
Kimi K3 deployment FAQ
How much GPU memory does Kimi K3 need?
The full checkpoint weight footprint is pending review. Minimum and recommended GPU configurations will be added after testing; the model name or active parameter count alone is not a memory requirement.
Has BenchGrid benchmarked Kimi K3?
Not yet. This profile contains publisher specifications and calculated weight-memory estimates. We do not currently publish measured latency, throughput, or cost per token for this model.
Where do these specifications come from?
The specifications are based on the official Moonshot AI model card linked on this page. Memory estimates use the stated total parameter count, including inactive experts for MoE models. Nominal model sizes are labeled with ~.
Explore your compute options.
Check available hardware, quotas, and current pricing with the provider. These links are not verified deployments or performance recommendations.
