Run Kimi-K3.1-Pro locally: hardware requirements
Kimi-K3.1-Pro needs about 1567 GB of GPU memory for one person at 32K context, and about 162 GB more for each extra person at 64K. For 80 users working as AI agents, plan on 4067 GB, which takes A cluster of about 24 × B200 across 3 servers.
Moonshot AI, China. From its files on Hugging Face, checked 2026-11-05.
Context per person: more context lets the model work with longer conversations and more files, and needs more GPU memory for every person. As a guide: 8K for short chat, 32K for chat over documents, 64K for AI agents, whose instructions, tools and files all count.
80 users means about 16 people working at the same moment. That share is a planning assumption; if you know yours, choose the total that matches it.
Work it out for your team: moonshotai/Kimi-K3.1-Pro
On Hugging FaceMoonshot AI, China · 2.8T parameters · License: other
What you need for 16 active users
A cluster of about 24 × B200 across 3 servers
The model is too large for one 8-card server, so its copy has to span servers. That needs an engine and network built for it.
The server
4 TB RAM
32 CPU cores
4 TB NVMe
Rules of thumb
- GPU memory in total
- 4067 GB
- Weights
- 1561 GB (published files)
- Conversation cache, all active
- 2581 GB
- Engine (vLLM)
- 3.0 GB
Ways to need less
- Use 32K context per personA cluster of about 17 × B200 across 3 servers
- Use 8K context per personA cluster of about 11 × B200 across 2 servers
- Use a smaller model: gpt-oss-120b1 × B200
- Use a smaller model: Mixtral-8x7B-Instruct-v0.11 × B200
Your options by card
80 users means about 16 using the model at any moment. The others are not waiting; they are simply doing something else. If more become active than a build handles, they queue. 17 cards too small for this model are left out.
No card on our list holds this model in one server.
These are memory limits. A card can hold more people than it can answer quickly, so test your own workload before you buy. More than one card in a server needs an engine that splits the model.
Kimi-K3.1-Pro by team size
AI agents, 1 user in 5 active at a time, 64K context, as published.
| Total users | At once | GPU memory | What you need |
|---|---|---|---|
| 20 | 4 | 2132 GB | A cluster of about 13 × B200 across 2 servers |
| 80 | 16 | 4067 GB | A cluster of about 24 × B200 across 3 servers |
| 120 | 24 | 5358 GB | A cluster of about 32 × B200 across 4 servers |
| 500 | 100 | 17616 GB | A cluster of about 104 × B200 across 13 servers |
Similar models
- Kimi-K3Moonshot AI · 2.8T · 4067 GB for 16 at once
- Kimi-K2.5Moonshot AI · 1.0T · 2277 GB for 16 at once
- Kimi-K2.6Moonshot AI · 1.0T · 2277 GB for 16 at once
- Kimi-K2.7-CodeMoonshot AI · 1.0T · 2277 GB for 16 at once
- Kimi-K2-InstructMoonshot AI · 1.0T · 2689 GB for 16 at once
- Qwen3.8-2.4T-A95BAlibaba (Qwen) · 2.4T · 4743 GB for 16 at once
How this is worked out
GPU memory = weights + a conversation cache for each person + the serving engine, from this model's files on Hugging Face. These are memory limits, not speed. The full method and its limits.
Questions
How much GPU memory does Kimi-K3.1-Pro need?
Kimi-K3.1-Pro needs about 1567 GB of GPU memory for one person at 32K context, and about 162 GB more for each extra person at 64K. For 80 users working as AI agents, plan on 4067 GB, which takes A cluster of about 24 × B200 across 3 servers.
How many active users can Kimi-K3.1-Pro handle on one 96 GB GPU?
Kimi-K3.1-Pro does not fit on one 96 GB card such as the RTX PRO 6000 Blackwell at 64K context. Compare larger cards or a smaller precision above.
Can Kimi-K3.1-Pro run on a 24 GB GPU such as the RTX 4090?
Not as published for one person at 32K context: it needs about 1567 GB. A 4-bit or 8-bit build may fit, if the model has one.
What license does Kimi-K3.1-Pro use?
Hugging Face lists its license as other. Check the license on the model's page before you use it.
Running a model on your own servers?
NeueCode is private autonomous intelligence for enterprise: it takes on your organization's tasks on your own GPUs, with nothing leaving your network, a person approving every action, and all of it on record. It runs each model on one card, so read the single-card figures above.
What you need to run NeueCode