What hardware do you need to run GLM-5.3-SLIM?

GLM-5.3-SLIM at UD-IQ2_M is 180 GB of weights, with a context ceiling of 1M tokens. This is GLM-5.3 with a quarter of its experts removed and the rest left at about 2.6 bits a weight: 192 of the original 256 routed experts survive in each of the 75 mixture-of-experts layers, picked by an architecture search, and the surviving weights are copied byte for byte out of Unsloth's two-bit build rather than quantised again. On the publisher's own runs it scores 92.7% on HumanEval, 84.8% on GPQA-Diamond and 71.4% on a 570-question MMLU subset; they did not run the unpruned build on the same harness, so the cost of cutting the experts is not measured there. Two losses stack, the missing experts and the two-bit weights, and the publisher reports a clear fall on Chinese knowledge questions. Nobody has placed this build on the index the rest of this site is ranked by. On the cheapest machine that runs it, a Mac Studio M5 Ultra, 256GB, it generates 23 tok/s at 32k of context, worked out from memory bandwidth.

How good is GLM-5.3-SLIM, really?

It has not been placed on the intelligence index yet. See the whole table.

Head to head with a model that needs much the same memory at 32k of context: vs Tencent Hy3.

What GLM-5.3-SLIM costs either way

Nobody rents GLM-5.3-SLIM by the token. The closest hosted match, GLM-5.3, costs $0.0699 per million input tokens and $1.4 per million output (OpenRouter, cheapest active endpoint, checked 2026-11-11). Buying a machine only beats that if you use it hard enough, for long enough, that the hardware price divides down below the rental bill. What a million tokens costs each way puts the two prices side by side.

Machines that run GLM-5.3-SLIM

The Mac Studio M5 Ultra, 256GB holds it to 64k, and it is the machine's memory that stops it there, not the model's 1M limit.

The table is the cheapest machine in each family, not the only one that runs it: 4 of the 56 machines on this site hold it at 32k. Within a family only memory decides, so that is every Mac Studio with 256GB or more. One table sets every machine against the others: price, the memory its GPU can use, what it holds and how long it takes to pay for itself.

One machine per family, cheapest first. Speeds are measured where a public benchmark exists and estimated from memory bandwidth otherwise; the calculator says which for any configuration. The longest context is the longest setting the calculator offers that the machine still holds this model at, cache included, and no machine is shown taking it past its own 1M limit. Each one opens the calculator on that machine at that length.

For scale, the calculator starts from 80 tok/s for a hosted API and times a local machine against it. The one above does not reach it, at 23 tok/s.

The cheapest machine that runs it is sold with 256 GB. There is a page on what 256 GB runs, machine by machine: it names the 2 machines here sold at that size, and the 51 current models the roomiest of them holds at 32k, this one among them.

Run GLM-5.3-SLIM at 64k on the Mac Studio M5 Ultra, 256GB

How much memory does GLM-5.3-SLIM need at 256k?

Every figure above is priced at 32k of context. At 256k, the longest the calculator offers, GLM-5.3-SLIM needs 204.8 GB rather than 183 GB. The weights are the same either way; the key-value cache goes from 3.1 GB to 25 GB.

No machine here that is still sold at a published price holds GLM-5.3-SLIM at 256k. One of the 37 machines here still sold at a published price holds it at 32k, and it is the Mac Studio M5 Ultra, 256GB at $10,799. Two machines here do hold it there: the Mac Studio M5 Ultra, 512GB, on sale with no published price; the Mac Studio M3 Ultra, 512GB, no longer sold, at the $9,499 it launched at.

The machines that miss GLM-5.3-SLIM, and what they run

Of the machines here with a published price that are still sold, every one but the Mac Studio M5 Ultra, 256GB misses GLM-5.3-SLIM, and none of them misses by a window: its 180 GB of weights are larger than the usable memory in every one of them, before a single token of key-value cache. The DGX Spark GB10 Grace Blackwell, 128GB comes nearest, 60.8 GB short at 4k, the shortest window on the list.

Three machines here do hold GLM-5.3-SLIM at 32k, and none of them is an answer to what to buy, which is the question the rest of this page asks: the Mac Studio M3 Ultra, 256GB and the Mac Studio M3 Ultra, 512GB, no longer sold, at the $7,099 and $9,499 they launched at, and the Mac Studio M5 Ultra, 512GB, on sale with no published price. They are sold with 256 GB and 512 GB of memory and hand a model 192 GB to 384 GB of it, against the 183 GB GLM-5.3-SLIM needs at 32k.

One machine per family, the largest configuration in each that still misses, nearest first. Short by is the weights plus the key-value cache at 4k measured against the memory that machine's GPU can address, which that machine's own page gives. The last column is the highest-scoring current model the machine holds at 32k.

Price the DGX Spark GB10 Grace Blackwell, 128GB on Qwen3.8 27B instead

The specifics

Parameters

563B, of which 40B are active per token

Quantisation

UD-IQ2_M

Weights on disk

180 GB

KV cache

3.1 GB at 32k context — Latent attention with a sparse-attention indexer beside it. All 78 blocks cache one 512-wide compressed vector plus 64 RoPE dimensions, 1,152 bytes a token; 21 of them compute their own index keys and cache 256 bytes more, and the other 57 share the keys of one that does. That comes to 93 KiB a token, the figure the publisher's own llama.cpp run reports. How weights and cache add up.

Maximum context

1M tokens

Licence

GLM-5.3 License

Sources

Scantist/GLM-5.3-SLIM-GGUF, zai-org/GLM-5.3 (config), zai-org/GLM-5.3, Artificial Analysis