Mistral Small 4 119B 2603 — Hardware Requirements & GPU Compatibility
ChatFunctionsMistral Small 4 119B is an MoE model from Mistral AI featuring 119 billion total parameters and roughly 6.5 billion active per token, utilizing 128 experts with 4 in use. This single checkpoint combines the company's Instruct, Reasoning (previously Magistral), and Devstral lines while accepting both text and images as input. Users can also adjust reasoning effort for each request. The card frames it well for chat assistants, coding tasks, agentic workflows with built-in function calling, and document analysis. Despite the few active parameters, all 119 billion weights must stay in memory, meaning it requires multi-GPU setups or a large unified-memory machine even when quantized. The card lists a context window of 256k tokens. It carries an Apache 2.0 license for both commercial and non-commercial use. Released as version 2603 in March 2026, it is positioned as the successor to Mistral Small 3, with the card reporting a 40 percent drop in end-to-end completion time during latency-optimized runs.
Mistral Small 4 119B 2603 runs on a 96 GB+ Mac or a workstation GPU
Best local pick → Mac Studio M5 Ultra (96 GB) · Q4_K_M · 81.6 GB · ~82 tok/s
Macs from 48 GB (IQ2_XXS · ~84 tok/s)
Apache 2.0
Not set up for it?
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Mistral Small 4 119B 2603
One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.
Recommended
ollama run hf.co/unsloth/Mistral-Small-4-119B-2603-GGUF:Q4_K_MQ4_K_M · needs ~81.6 GB · 16K context
Download: UD-Q4_K_M/Mistral-Small-4-119B-2603-UD-Q4_K_M-00001-of-00003.gguf73.76 GB + 2 partsunsloth
Ways to runOther ways
each opens step-by-stepRun in cloud
Your own private copy · Q4_K_M · 96 GB GPU
- The easiest way to try this model
- Running in a few minutes, easy setup
- Chat right here or connect any app
$5+ bonus for new users
Runs on H200 (141 GB) (59 GB headroom) · Q4_K_M
- Generation speed
- ~297 tok/s
- generation speed
- Cost per 1M output tokens
- $3.36
- per 1M output tokens
How Much VRAM Does Mistral Small 4 119B 2603 Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 52.3 GB | 70.4 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Does not fit: needs 60.7 GB | |
| 53.8 GB | 71.9 GB | 3-bit small — 3-bit small quantization | Does not fit: needs 62.2 GB | |
| 59.7 GB | 77.8 GB | 3-bit medium — 3-bit medium quantization | Does not fit: needs 68.2 GB | |
| 61.2 GB | 79.3 GB | 4-bit · legacy — 4-bit legacy quantization | Does not fit: needs 69.7 GB | |
| 73.2 GB | 91.3 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Does not fit: needs 81.6 GB | |
| 86.6 GB | 104.7 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Does not fit: needs 95 GB | |
| 100.0 GB | 118.1 GB | 6-bit, very good — 6-bit quantization, very good quality | Does not fit: needs 108.5 GB | |
| 120.9 GB | 139 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Does not fit: needs 129.4 GB |
Selected Q4_K_M
4-bit medium quantization81.6GBat 16K context
- Weights
- 71.6 GB
- KV cache
- 9.7 GB
- Overhead
- 0.3 GB
Doesn't fit RTX 4090 (reference) · needs 57.6 GB more
Or .
Which Hardware Can Run Mistral Small 4 119B 2603?
Q4_K_M · 73.2 GBMistral Small 4 119B 2603 (Q4_K_M) requires 73.2 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 96+ GB is recommended. Using the full 1049K context window can add up to 617.3 GB, bringing total usage to 690.4 GB. No consumer GPU has enough memory. 34 devices with unified memory can run Mistral Small 4 119B 2603, including Mac Studio M5 Ultra (256 GB), Mac Studio M5 Ultra (512 GB), Mac Studio M5 Ultra (96 GB).
48 GB
48 GB tier →Q4_K_M · best fit IQ2_S (47.3 GB)
2 Mac classes · 1 device family
M5 Max48 GB~80Simulate output on Mac Studio M5 Max (48 GB)M5 Pro48 GB~56Simulate output on Mac Mini M5 Pro (48 GB)Snapdragon X2 Elite Extreme (X2E-96-100)~52Simulate output on Snapdragon X2 Elite Extreme Copilot+ PC96 GB
fits · 14.4 GB free · up to 32K ctx
2 Mac classes
M5 Ultra96 GB~82Simulate output on Mac Studio M5 Ultra (96 GB)M3 Max*96 GB~36Simulate output on MacBook Pro 14" M3 Max (96 GB)
By VRAM tier · typical GPUs and Macs at Q4_K_M
How good is Mistral Small 4 119B 2603?
llmrun Score blends public benchmark boards into one number; 100 = the average open model.
Not enough data yet for an llmrun Score.
A Score needs results on enough independent boards. How the Composite Score is built
Benchmarks
Here are all 3 boards that have results for this model, sorted by their weight in the llmrun Score; badges show who ran the test.
Alternatives to Mistral Small 4 119B 2603
Mistral family · 37 models →Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| GPT OSS 120BOpenAI | 97.5 | 70.8 GB | —needs > 24 GB | same size |
| C4ai Command A 03 2025Cohere | 79.8 | 78.7 GB | —needs > 24 GB | same size |
| Ornith 1.5 122Bornith-ai | — | 75.5 GB | —needs > 24 GB | same size |
| NVIDIA Nemotron 3.5 Super VL 120B A12B BF16NVIDIA | — | 87.7 GB | —needs > 24 GB | same size |
| Ling 3.0 Flash VLInclusion AI | — | 76 GB | —needs > 24 GB | same size |
| Qwen1.5 110BAlibaba | — | 72.4 GB | —needs > 24 GB | same size |
Specifications
- Publisher
- Mistral AI
- Family
- Mistral
- Parameters
- 119.4B
- Active parameters
- 7.1B per token
- Architecture
- Mistral3ForConditionalGeneration
- Context Length
- 1,048,576 tokens
- Vocabulary Size
- 131,072
- Release Date
- 2026-01-23
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Mistral Small 4 119B 2603 need?
Mistral Small 4 119B 2603 requires 73.2 GB of VRAM at Q4_K_M, or 240.3 GB at BF16. Full 1049K context adds up to 617.3 GB (690.4 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 119.4B × 4.8 bits ÷ 8 = 71.6 GB
KV Cache + Overhead ≈ 1.6 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
KV Cache + Overhead ≈ 618.8 GB (at full 1049K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 5090 run Mistral Small 4 119B 2603?
No — Mistral Small 4 119B 2603 requires at least 34.3 GB at IQ2_XXS, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.
What's the best quantization for Mistral Small 4 119B 2603?
For Mistral Small 4 119B 2603, Q4_K_M (73.2 GB) offers the best balance of quality and VRAM usage. Q4_K_L (74.6 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 34.3 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Mistral Small 4 119B 2603 on a Mac?
Yes — Mac Studio M5 Max (48 GB) and 58 other Macs can run Mistral Small 4 119B 2603, the smallest at IQ2_XXS. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 73.2 GB of usable unified memory (RAM minus macOS overhead), from Mac Studio M5 Ultra (96 GB).
Can I run Mistral Small 4 119B 2603 locally?
Yes — Mistral Small 4 119B 2603 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Mistral Small 4 119B 2603?
At Q4_K_M, Mistral Small 4 119B 2603 can reach ~123 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)
Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.
Example: NVIDIA B200 → 4.3 GB active ÷ (8000 × 0.65) = 0.83 ms, plus 36 layers × 0.055 ms = 1.98 ms, so 1000 ÷ 2.81 ms = ~356 tok/s
Estimated speed at Q4_K_M (73.2 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Mistral Small 4 119B 2603?
At Q4_K_M, the download is about 71.64 GB. The full-precision BF16 version is 238.80 GB. The smallest option (IQ2_XXS) is 32.84 GB.
Which GPUs can run Mistral Small 4 119B 2603?
No single consumer GPU has enough VRAM to run Mistral Small 4 119B 2603 at Q4_K_M (73.2 GB). Multi-GPU or professional hardware is required.
Which devices can run Mistral Small 4 119B 2603?
34 devices with unified memory can run Mistral Small 4 119B 2603 at Q4_K_M (73.2 GB), including ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB), GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.