Architecture truth card — Kimi-K3.1-Pro
← Open the interactive VRAM & cost calculator on the homepage
Exact VRAM matrix — 3 precisions × 3 context lengths
| Precision | 8192 tokens ctx | 32768 tokens ctx | 131072 tokens ctx |
|---|---|---|---|
| FP16/BF16 16-bit | 5178.36 / 0.82 / 260.42 Total 5439.59 GiB | 5178.36 / 3.27 / 260.42 Total 5442.05 GiB | 5178.36 / 13.08 / 260.42 Total 5451.86 GiB |
| FP8/INT8 8-bit | 2589.18 / 0.82 / 130.96 Total 2720.96 GiB | 2589.18 / 3.27 / 130.96 Total 2723.41 GiB | 2589.18 / 13.08 / 130.96 Total 2733.22 GiB |
| INT4/AWQ 4.5-bit | 1456.41 / 0.82 / 74.32 Total 1531.55 GiB | 1456.41 / 3.27 / 74.32 Total 1534.00 GiB | 1456.41 / 13.08 / 74.32 Total 1543.81 GiB |
Each cell: Weights GiB / GQA KV Cache GiB / Runtime Overhead GiB; bold line = Total VRAM GiB. Concurrency = 1, FP16 KV cache (2 bytes), overhead = 1.5 + 0.05 × weights GiB.
Terminal speed-math — no install needed
Re-run this exact sizing from your own terminal:
curl -fsSL https://gpuvram.pages.dev/vram-cli.py | python3 - kimi-k3-1-pro --ctx 32768
Live cheapest GPU configurations
| Scenario | Cheapest fitting rig (top 3) | Total VRAM | Lowest $/hr | 730 h/mo | Deploy |
|---|---|---|---|---|---|
| INT4 @ 32K ctx | 8x MI300X 192GB | 1536 GiB | $4.00 | $2920.00 | Deploy on RunPodCheck Vast.ai Verified |
| INT4 @ 32K ctx | 8x B300 288GB | 2304 GiB | $55.52 | $40529.60 | Deploy on RunPodCheck Vast.ai Verified |
What over-provisioning costs you
Kimi-K3.1-Pro at INT4 / 32K context fits the cheapest rig (8x MI300X 192GB) for $2920.00/mo (730 h). Scaling the same INT4 workload to 131K context pushes the KV cache past that rig: the cheapest 131K-capable rig (8x B300 288GB) bills $40529.60/mo — a concrete $40529.60 instead of $2920.00 gap you pay monthly for headroom the 32K rig physically lacks.