Our Kimi K3 spotlight ended on a wall: the largest open model ever announced, and a single 512GB Mac Studio cannot hold it at any quant. The obvious next question, the one filling our inbox, is "fine, so what should I run it on?" This is the answer, ranked, with the projected memory and cost math for each option.
One caveat up front, same as the spotlight: the weights are not downloadable yet (Moonshot says by July 27, 2026), so nobody has run K3 locally and there are no community quants. Every size below is projected from the confirmed 2.8-trillion-parameter figure and the real, measured file sizes of the previous model, Kimi K2. We will replace projections with measured numbers the moment the weights land.
The answer most people do not want to hear
For almost everyone, the optimal hardware for Kimi K3 is none. It is a 2.8-trillion-parameter model, which is a datacenter model, not a desktop one. Moonshot's own API costs $3 per million input tokens and $15 per million output ($0.30 cached), and no machine you can buy will beat that on cost-per-token for a model this size. Buy hardware for K3 only if you specifically need to self-host it: privacy, data control, or offline use. If that is not you, the buy-vs-rent-vs-API math ends at "use the API," and you can stop reading.
Still here? Then you want to self-host, and the rest of this is for you.
The two numbers that decide everything
A Mixture-of-Experts model has two parameter counts, and they do different jobs. Total parameters decide whether it fits in your memory. Active parameters decide how fast it decodes, because generation is memory-bandwidth-bound: each token only reads the experts that fire. Our MoE explainer covers the mechanism.
For K3, total is a firm 2.8T. Active is the catch: Moonshot only published the routing ("16 of 896 experts" fire per token) and gave no per-expert size, so the active count in billions is not officially known. The ~50B active figure you will see quoted is an estimate from that 16/896 ratio, not a confirmed number. Treat every tok/s below as a projection with that asterisk.
The trap this creates: "only 50B active" does not mean K3 runs on a 50B-class card. The router picks different experts every single token, so you must keep the entire model in fast memory or pay a catastrophic streaming penalty. Total params set what fits; active params only set the speed once it does.
Projected footprint
Kimi K2 (1T total) ships real GGUF files at about 587GB for Q4_K_M and 1.09TB for Q8, which is where we documented the 594GB reality for K2.7. Scale by 2.8x, and cross-check against the file-size rule (bytes = parameters x bits-per-weight / 8):
| Quant | Projected K3 weights (2.8T) |
|---|---|
| Q8 | ~2.8 to 3.0 TB |
| Q4_K_M | ~1.6 to 1.7 TB |
| Heavy IQ2 | ~0.9 to 1.0 TB |
Weights only. Add ~10-15% runtime overhead. KV cache stays small (tens of GB even at 1M context) because K3 uses Kimi Delta Attention, a linear-attention scheme.
What actually holds it, ranked
The rule is simple: your fast memory has to exceed the quant size above, plus overhead. Here is where the common machines land.
| Hardware (usable memory) | Runs K3? |
|---|---|
| RTX 5090 32GB, or 2-4x RTX 3090 (48-96GB) | No, at any quant (off by 8 to 30x) |
| Strix Halo 128GB unified | No (IQ2 needs ~0.9TB, about 8x short) |
| Mac Studio M3 Ultra 512GB (single) | No, even at IQ2 (~0.9TB > 512GB) |
| 8x H100 80GB (640GB) | No, even for IQ2 |
| 8x H200 141GB (1,128GB) | IQ2 fits with KV room; full Q4 does not |
| 2-3x Mac Studio 512GB, clustered | ~1 to 1.5TB unified, fits IQ2 or a low Q3 |
The single 512GB Mac line is the reversal from K2: one Mac runs K2 at Q2 comfortably, but cannot hold K3 at any quant.
The optimal box, by goal
Best single-node self-host: an 8x NVIDIA H200 server (1,128GB). It fits a heavy IQ2/Q3 build with headroom, and its roughly 4.8 TB/s of bandwidth per card would decode fastest. If the ~50B active estimate holds, that points to somewhere around 80 to 130 tok/s, an order of magnitude past anything prosumer. The cost is the catch: $250,000-plus to own, or rent it by the hour. A full Q4 (~1.7TB) needs about twice that, a 16x H200 setup or a B200/GB200-class node.
Cheapest that genuinely runs it: 2 to 3 Mac Studio M3 Ultra 512GB machines, clustered with a distributed runtime (EXO or MLX) to reach 1 to 1.5TB of unified memory for an IQ2 or low-Q3 quant. Roughly $20,000 to $30,000. It works, but it is slow: each Mac moves 819 GB/s, and once you subtract the Thunderbolt interconnect overhead of splitting a model across boxes, expect something in the 10 to 20 tok/s range. This is the only path that does not require a rack.
Smartest for most self-hosters: rent. An H200 or B200 node on RunPod, Vast, or Lambda, hourly, gives you the fast option with no capital and nothing to babysit. For a model you reach for occasionally, renting the big node when you need it beats owning a $250k box that a much smaller model could serve the rest of the time. This is usually the real answer, and it is exactly the tradeoff our cost calculator is built to price.
A note on speed, since bandwidth rules here
Decode speed on any of these is set by memory bandwidth divided by the bytes read per token, and the bytes read are set by the active parameters, not the 2.8T total. As we cover in Bandwidth, Not TFLOPS, this is why an H200 node (4.8 TB/s) would leave a Mac cluster (819 GB/s per unit, minus interconnect) far behind on tokens per second even when both technically fit the model. If Moonshot kept the active count modest, K3 could decode at a genuinely usable clip on the right box. We just cannot confirm the number until the weights ship.
The verdict
For nearly everyone: use the API, or rent a big GPU node by the hour. Self-hosting Kimi K3 is a five-figure-hardware, cluster-or-datacenter proposition, and the realistic ceiling on "prosumer" is a $20k to $30k Mac Studio cluster running the most brutal quant slowly. If you have the budget and the need, an 8x H200 node is the cleanest single-box answer; if you have the need but not $250k, rent that same class of node. And if you are weighing any of this seriously, wait for July 27: the first real quants and owner token rates will replace every projection here, and we will update this piece with measured numbers the day they land.
Sources and how we researched this
- Specs (2.8T total, 16 of 896 experts, 1M context, Kimi Delta Attention, MXFP4 training): Moonshot AI's Kimi K3 announcement.
- Measured footprint baseline: the real Kimi K2 GGUF sizes (Q4 ~587GB, Q8 ~1.09TB), scaled by 2.8x and cross-checked with the params x bits / 8 rule.
- Weights-not-out status (release by July 27, 2026, no public download yet) confirmed via Moonshot's blog and independent reporting.
- Hardware memory and bandwidth figures are vendor specs (NVIDIA H100/H200, Apple M3 Ultra). We have not run K3 first-hand; all K3 sizes and speeds are projected from the above, not measured.
Related: Kimi K3: the 2.8T spotlight (part 1) · Why active parameters decide speed · Mac Studio M3 Ultra for local AI · Kimi K2.7 and the 594GB reality
