What hardware do you need to run Mistral Small 4 (119B-2603)?

Mistral Small 4 (119B-2603) at Q4_K_M is 74 GB of weights, with a context ceiling of 1M tokens. Despite the name it is a 119B MoE. On the cheapest machine that runs it, a Framework Desktop, 128GB, it generates 32 tok/s at 32k of context, worked out from memory bandwidth.

How good is Mistral Small 4 (119B-2603), really?

On the Artificial Analysis Intelligence Index v4.3 it scores 11 (reasoning mode; 9 without), which puts it in the Below every hosted tier band. Fine for simple, well-specified tasks. Noticeably less capable than anything the big labs sell today. See the whole table.

Head to head with its neighbours on the leaderboard: vs GLM-4.5-Air above it, vs Qwen3 30B-A3B Instruct 2507 below. One more model needs much the same memory at 32k of context: vs Ling 3.0 flash.

What Mistral Small 4 (119B-2603) costs either way

Renting the same model costs $0.15 per million input tokens and $0.6 per million output (OpenRouter, cheapest active endpoint, checked 2026-10-29). 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 Mistral Small 4 (119B-2603)

Every machine here holds it to 256k, the longest the calculator offers, so the choice between them is speed and price rather than how much you can keep in the window.

The table is the cheapest machine in each family, not the only one that runs it: 17 of the 56 machines on this site hold it at 32k. Within a family only memory decides, so that is every NVIDIA card with 96GB or more, every Mac Studio, Strix Halo box and MacBook Pro with 128GB or more and the DGX Spark, 128GB. 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. 1 of the 5 above reaches it, and the slowest is 32 tok/s.

The cheapest machine that runs it is sold with 128 GB. There is a page on what 128 GB runs, machine by machine: it names the 11 machines here sold at that size, and the 39 current models the roomiest of them holds at 32k, this one among them.

It fits in less memory than that: the smallest size here that runs it at 32k is 96 GB, on the RTX PRO 6000 Blackwell, 96GB at $18,000, card only, which hands a model 95 GB.

Run Mistral Small 4 (119B-2603) at 256k on the Strix Halo Framework Desktop, 128GB

How much memory does Mistral Small 4 (119B-2603) need at 256k?

Every figure above is priced at 32k of context. At 256k, the longest the calculator offers, Mistral Small 4 (119B-2603) needs 79.8 GB rather than 74.5 GB. The weights are the same either way; the key-value cache goes from 0.8 GB to 6 GB.

The window costs nothing in hardware: the same 12 of the 37 machines here still sold at a published price hold Mistral Small 4 (119B-2603) at both lengths, and the cheapest of them is the Framework Desktop, 128GB at $3,449.

On the Framework Desktop, 128GB it runs 32 tok/s estimated at 32k and 14 tok/s estimated at 256k. Every token generated reads the whole cache back, so a longer window is slower as well as bigger.

The specifics

Parameters

119.4B, of which 6B are active per token

Quantisation

Q4_K_M

Weights on disk

74 GB

KV cache

0.8 GB at 32k context — Latent attention: the cache holds a 256-wide compressed vector plus 64 RoPE dimensions per layer, 640 bytes a token. The 32 KV heads in the config are not a cache width — reading them that way overstates the memory 25-fold. How weights and cache add up.

Maximum context

1M tokens

Licence

Apache 2.0

Sources

unsloth/Mistral-Small-4-119B-2603-GGUF, mistralai/Mistral-Small-4-119B-2603