Mistral AI · dense · Modified MIT
What runs Mistral Medium 3.5 128B
Dense 128B model that replaces Mistral Medium 3.1 and Magistral as Mistral's flagship open release. Being dense, its speed is set by memory bandwidth across every parameter.
- Parameters
- 128B
- 4-bit file
- 69.8 GB
- Q4_K_M
- Needs at 8k context
- 74.0 GB
- Longest context
- 262,144 tokens
Context memory: 360,448 bytes per token. Standard GQA: 2 x 88 layers x 8 KV heads x 128 head dim x 2 bytes (fp16). How context memory works.
Measured on this hardware
Machines that hold the 4-bit file with 8,192 tokens of context and have published runs on the same chip or cards, fastest first. Median of single-user runs with plain decoding.
No published run on a machine that fits it.
Estimated, not measured
These fit but nobody has published a run we could match. The figure is memory bandwidth divided by the bytes read per token, scaled by the share of that ceiling measured runs reach. How estimates work.
| Machine | Fit | Price | Generation |
|---|---|---|---|
| Dell Pro Precision with GB300NVIDIA workstation | Fits | ~$159,283 | ~64–77 tok/sestimate |
| Dell Pro Max Tower T2 with NVIDIA RTX PRO 6000 BlackwellNVIDIA workstation | Fits | ~$13,909 | ~12–18 tok/sestimate |
| 1x RTX PRO 6000 BlackwellSingle GPU | Fits | ~$20,833 | ~12–18 tok/sestimate |
| Apple Mac Studio (M5 Ultra, 64-core GPU)Apple unified memory | Fitsassumed limitfrom 256 GB | ~$5,499–$9,499 | ~12–15 tok/sestimate |
| Apple Mac Studio (M5 Ultra, 80-core GPU)Apple unified memory | Fitsassumed limitfrom 256 GB | ~$6,799–$10,799 | ~12–15 tok/sestimate |
| 4x Instinct MI50 32GB (used) on EPYC 7003Multi-GPU | Fits | ~$1,998 + unpriced parts | ~6.7–11 tok/sestimate |
| 4x RTX 3090 (used) on EPYC 7003, open frameMulti-GPU | Fits | ~$2,178 + unpriced parts | ~6.1–9.6 tok/sestimate |
| RTX 5090 + EPYC 9355P with 768 GB DDR5 (MoE offload)Hybrid GPU + RAM | GPU + RAM offloadassumed limit | ~$3,763 + unpriced parts | ~5.4–8.6 tok/sestimate, part in RAM |
| Apple Mac Studio (M5 Max, 40-core GPU)Apple unified memory | Fitsassumed limitfrom 128 GB | ~$3,099–$5,099 | ~6.0–7.5 tok/sestimate |
| Apple MacBook Pro 14-inch (M5 Max, 40-core GPU)Apple unified memory | Fitsassumed limitfrom 128 GB | ~$4,699–$6,699 | ~6.0–7.5 tok/sestimate |
| Apple MacBook Pro 16-inch (M5 Max, 40-core GPU)Apple unified memory | Fitsassumed limitfrom 128 GB | ~$4,999–$6,999 | ~6.0–7.5 tok/sestimate |
| tiny corp tinybox red flashGPU desktop (prebuilt) | Fits | ~$22,000 | ~4.2–6.6 tok/sestimate |
| Framework Desktop (Ryzen AI Max+ PRO 495)AMD unified memory | Fitsassumed limit | ~$6,799 | ~2.5–3.0 tok/sestimate |
| NVIDIA DGX SparkNVIDIA GB10 | Fitsassumed limit | ~$4,699 | ~2.5–3.0 tok/sestimate |
| Dell Pro Precision with GB10NVIDIA GB10 | Fitsassumed limit | ~$5,523 | ~2.5–3.0 tok/sestimate |
| Framework Desktop (Ryzen AI Max+ 395)AMD unified memory | Fitsfrom 128 GB | ~$1,959–$3,449 | ~2.3–2.8 tok/sestimate |
| GMKtec EVO-X2 (Ryzen AI Max+ 395)AMD unified memory | Fitsfrom 128 GB | ~$2,600–$4,000 | ~2.3–2.8 tok/sestimate |
| Beelink GTR9 Pro (Ryzen AI Max+ 395)AMD unified memory | Fits | ~$4,699 | ~2.3–2.8 tok/sestimate |
| Minisforum MS-S1 MAX (Ryzen AI Max+ 395)AMD unified memory | Fits | ~$4,899 | ~2.3–2.8 tok/sestimate |
| Corsair AI Workstation 300 (Ryzen AI Max+ 395)AMD unified memory | Fits | ~$4,800 | ~2.3–2.8 tok/sestimate |
| HP Z2 Mini G1a (Ryzen AI Max+ PRO 395)AMD unified memory | Fits | ~$4,999 | ~2.3–2.8 tok/sestimate |
| RTX 3090 (used) + 128 GB DDR5 on AM5 (MoE offload)Hybrid GPU + RAM | GPU + RAM offloadassumed limit | ~$4,833 + unpriced parts | ~0.9–1.4 tok/sestimate, part in RAM |
Fits, no speed figure
Too large for
Apple Mac Studio (M5 Max, 32-core GPU), Apple Mac mini (M5 Pro, 16-core GPU), Apple Mac mini (M5 Pro, 20-core GPU), Apple MacBook Pro 14-inch (M5 Max, 32-core GPU), Apple MacBook Pro 16-inch (M5 Max, 32-core GPU), Framework Desktop (Ryzen AI Max 385), 1x RTX 3090 (used), 2x RTX 3090 (used), 1x RTX 5090, 2x RTX 5090, 2x Radeon RX 7900 XTX, 2x Intel Arc Pro B60. A smaller quant may fit: try the calculator.
Files
| Quant | Format | Size | Repo |
|---|---|---|---|
| Q4_K_M | GGUF | 69.8 GB | unsloth/Mistral-Medium-3.5-128B-GGUF |
| 4bit | MLX | 72.6 GB | mlx-community/Mistral-Medium-3.5-128B-4bit |
| Q5_K_M | GGUF | 82.3 GB | unsloth/Mistral-Medium-3.5-128B-GGUF |
| Q6_K | GGUF | 95.5 GB | unsloth/Mistral-Medium-3.5-128B-GGUF |
| Q8_0 | GGUF | 123.7 GB | unsloth/Mistral-Medium-3.5-128B-GGUF |
Sources
- huggingface.co: model card: params, context, license, attention layout, description
- huggingface.co: downloads (last 30 days), license tag, safetensors param count
- huggingface.co: layer count, KV heads and head size
- huggingface.co: GGUF file sizes (shards summed)
- huggingface.co: MLX 4bit safetensors sizes