Runs on ≤16 GB · Hugging Face ↗
Coding index 14.4 (Artificial Analysis)
Mistral's dense 14B in the edge-oriented Ministral line — a compact generalist with solid instruction-following.
HumanEval pass@1 63% (19/30), via local GPU (Ollama, Q4_K_M). Measured by us on this hardware, as a rough sanity check — HumanEval is a different, easier, partly-contaminated benchmark than Artificial Analysis’ composite, so it is not comparable to the coding-index column.
BigCodeBench-Hard pass@1 14% (17/121), via local GPU (Ollama, Q4_K_M). The brutal counterpart to HumanEval — where HumanEval saturates near the top, BCB-Hard spreads the field, so this is the number that actually separates coding ability.
~30 tok/s on RTX 4060 Ti, measured by us.
Real GGUF file sizes = weight VRAM.
| Quantization | Size |
|---|---|
| Q3_K_M | 6.7 GB |
| IQ4_XS | 7.4 GB |
| Q4_0 | 7.8 GB |
| Q4_K_S | 7.8 GB |
| Q4_K_M | 8.2 GB |
| Q4 | 8.4 GB |
| Q4_1 | 8.6 GB |
| Q5_K_M | 9.6 GB |
| Q8_0 | 14.4 GB |
| Q2_K | 16.2 GB |
| Q8 | 17.1 GB |
| Q6_K | 23.2 GB |
| BF16 | 27.0 GB |
~8 GB at Q4, comfortable on a 16 GB card with long context. Respect the Mistral v3 tokenizer / chat template.