Mistral AI
Mistral Small 4 (26.03)
119 B params (6.5 B active) · 256k context · Open weights
Released 2026-03-16
Details
API pricing
Running it locally
Memory needed for 119 B parameters (mixture of experts: 6.5 B active per token, but all 119 B must be held in memory) at 8,192 tokens of context. Architecture not recorded, so a flat 1.0 GB KV allowance is used. Estimates, not measurements.
| Format | Weights | KV cache | Overhead | Total |
|---|---|---|---|---|
| Q4 K_M | 72.7 GB | 1.0 GB | 4.1 GB | 77.9 GB |
| Q6 K | 97.6 GB | 1.0 GB | 5.4 GB | 104.0 GB |
| Q8 0 | 126.4 GB | 1.0 GB | 6.8 GB | 134.3 GB |
| FP16 | 238.0 GB | 1.0 GB | 12.4 GB | 251.4 GB |
Catalogued hardware that fits at Q4
- GMKtec EVO-X3 (Ryzen AI Max+ 395)comfortable
- Beelink GTR9 Pro (Ryzen AI Max+ 395)comfortable
- OneXPlayer OneXStation i1comfortable
- Framework Desktopcomfortable
Mini PCs reserve 8.0 GB for the operating system. “Comfortable” keeps at least 10% memory spare.
Benchmarks
No benchmarks for this model yet
We only publish results we can evidence, so there is nothing here yet and we have not filled the gap with estimates. Here is how we will test and label results.