# Can you run Kimi K3 locally? Not on a desk, and Moonshot never said you could (slim)

> Vlad Cealicu on what it takes to run Moonshot's Kimi K3, 2.8 trillion parameters with 104 billion active, yourself. The official weights are 1,561 GB, the smallest community file is 466 GB, the smallest setups shown serving it are eight-GPU B300 nodes, and what the Mac, mini PC, DGX Station and CPU server options would cost against the API.

- Full: https://www.anchorterminal.com/blog/run-kimi-k3-locally.md (~3,150 tokens) · this version ~1,530 tokens · JSON https://www.anchorterminal.com/blog/run-kimi-k3-locally.json · canonical https://www.anchorterminal.com/blog/run-kimi-k3-locally
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-10-25
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

By Vlad Cealicu, 25 October 2026. 1,094 words, about 5 min to read.

When Meta opened the weights of Muse Spark Open, its announcement said two Mac Studios were "the best way to run a powerful model on your own desktop", and I wrote that [the vast majority of us can't](/blog/run-muse-spark-open-locally).

This is the second post in a series on what it takes to run an open model yourself.

## How big is Kimi K3?

2.8 trillion parameters, with 104 billion active for each token.

The part that matters at home is that Moonshot trained it to run in 4-bit from the fine-tuning stage on, so the official weights are already 4-bit, and they're still 1,561 GB [[2]](#ref-2).

- 1,561 GB: the official weights, already 4-bit
- 594 GB: Unsloth's recommended 1-bit file, about 610 GB of memory to run
- 466 GB: the smallest file anyone has published, quality unknown

Kimi K3's weights, 24 October 2026. Sources [2] [3].

## What do the smaller files cost you?

Unsloth publishes GGUF files from 466 GB up.

Size in GB

| File | GB |
| --- | --- |
| 1-bit, smallest | 466 |
| 1-bit, recommended | 594 |
| 2-bit | 861 |
| 4-bit | 1509 |
| 8-bit, lossless | 1561 |

Each series has a zero-based scale.

## What hardware would hold it?

These are my sums, using about 120 GiB of usable GPU memory on a 128GB machine as in the Muse post, which is about 129 GB [[5]](#ref-5).

For the full 1,561 GB, seven 256GB Mac Studios would hold it, about $66,500 to $75,600 [[6]](#ref-6).

For Unsloth's 594 GB 1-bit file, it's three 256GB Mac Studios at about $28,500 to $32,400, five Minisforum MS-S1 Max at about $19,600, or five DGX Sparks at about $34,750 [[5]](#ref-5) [[6]](#ref-6) [[9]](#ref-9) [[10]](#ref-10).

What holding Kimi K3 costs, US dollars, lower prices

| Setup | US dollars |
| --- | --- |
| One DGX Station, 1-bit | 94930 |
| Seven 256GB Mac Studios, full weights | 66500 |
| Five DGX Sparks, 1-bit | 34750 |
| Three 256GB Mac Studios, 1-bit | 28500 |
| Five MS-S1 Max, 1-bit | 19600 |
| 2TB of DDR4 for a CPU server, the RAM alone | 10000 |

Each series has a zero-based scale.

## How fast would it be?

Nobody has published a good number on hardware a person owns.

In theory a CPU server could do better.

- 1: token a second, a MacBook Pro streaming from four SSDs
- 0.1: token a second, a rented server with four A100s
- 122: tokens a second across 16 users, eight B300s

Kimi K3's published speeds.

The 122 is imec's, on an eight-GPU B300 server running 16 coding agents at once, about 7.6 tokens a second each, and the same post says K3 doesn't fit on an eight-GPU B200 server [[13]](#ref-13).

## Is it cheaper to use the API?

By a long way.

## What I'd do

Use the API, and say thank you to Moonshot for publishing the weights anyway, because they let other providers compete on price and let researchers look inside.

If your data can't leave the building, K3 is a server job, not a desk job.

Kimi K3 now has [its own listing](/tools/kimi-k3) graded by our open-weight checklist, with its builds, the memory each needs and the command that starts it, and [a page comparing it with GLM-5.3-Flash](/compare/glm-5-3-flash-vs-kimi-k3).

---

More in [Run it locally](https://www.anchorterminal.com/blog/run-it-locally/index.md) · Previous model [Can you run Muse Spark Open locally? Hardware, RAM and cost](https://www.anchorterminal.com/blog/run-muse-spark-open-locally.md)

- Older: [The Log, issue 3: found, leaked, let in](https://www.anchorterminal.com/blog/the-log/issue-3-found-leaked-let-in.md)
- All posts: https://www.anchorterminal.com/blog/index.md

## References

Read on 24 October 2026.

[1] Moonshot AI, Kimi K3, on the Kimi blog.

[2] Moonshot AI, Kimi-K3 on Hugging Face, with its config and licence.

[3] Unsloth, Kimi K3 guide and unsloth/Kimi-K3-GGUF.

[4] mlx-community, Kimi-K3-mlx-reap160-2bit.

[5] Anchor Terminal, can you run Muse Spark Open locally, 15 October 2026. Source for counting about 120 GiB of usable memory on a 128GB machine.

[6] Daring Fireball, configurations and pricing, 25 August 2026, and Apple's Mac Studio specs and store pages.

[7] DatacenterDisk, server RAM prices, updated 23 October 2026. Source for DDR4 at $4.92 a GB for the cheapest, DDR4-2400 sticks, and $9.06 a GB at the median.

[8] Tom's Hardware, 23 August 2026, and buyzero.de, read 24 October 2026. Source for an Exxact DGX Station listed at $94,930 in August, buyzero listing nine models from €104,800 on 24 October, 252 GB of HBM3e plus 496 GB of LPDDR5x, and two Stations linking into about 1.5 TB.

[9] Minisforum US store, MS-S1 Max, read 24 October 2026. Source for $3,919 for the 128GB model.

[10] The Register, 2 October 2026. Source for the 128GB DGX Spark at $6,950. https://www.theregister.com/systems/2026/10/02/nvidia-debuts-4999-dgx-spark-with-half-the-ram-and-storage-amid-memory-crunch/5300622

[11] konstantinnikol, Kimi-K3 discussion 185 on Hugging Face, 17 September 2026. Source for 0.96 to 1.00 tokens a second on a 128GB MacBook Pro M5 Max streaming the 4-bit experts from four SSDs, with an engine from the poster's own company.

[12] ComputingForGeeks, run Kimi K3 locally, updated 10 October 2026. Source for 0.10 tokens a second on a rented vast.ai server with four A100s and 2TB of RAM, GPU use between 0 and 1 per cent and its NUMA setup.

[13] imec AI stack team, GPU self-hosting, updated 29 July 2026. Source for 122 tokens a second across 16 sessions on eight B300s and K3 not fitting an eight-GPU B200 node.

[14] NVIDIA, Kimi-K3-NVFP4 on Hugging Face.

[15] OpenRouter, Kimi K3 endpoints.

[16] Moonshot AI, Kimi K3.1 Pro tech blog, on the Kimi blog.

[17] The Next Web, Moonshot releases Kimi K3.1 Pro and says it cut its hacking skills, 23 October 2026. Source for the release on Friday 23 October, as The Next Web reported it.