Model benchmarks
Ministral-3-8B-Instruct-2512-Q4_K_M local LLM performance
As of October 2026, Ministral-3-8B-Instruct-2512-Q4_K_M runs at up to 35.4 tok/s for local inference (best of 6 community benchmark runs across 1 GPU).
Model size
8B
Peak speed
35.4 tok/s
Average speed
34.9 tok/s
Avg PP
276.9 tok/s
Min memory
7.7 GB
Max context
32,768 tokens
Avg output / run
6,968 tokens
Avg runtime / run
3m 37s
Avg quality
64.4
Benchmark runs
6
GPUs tested
1
Quality by task
Average LLM-judged quality (0–100) with the run-to-run spread shown as a P5–P95 band, overall and for each benchmark task, across all 6 runs. The low and high columns show how much the judge’s score varies between runs, and need at least two runs to display.
| Task | P5 (low) | Avg | P95 (high) |
|---|---|---|---|
| Overall | 61.2 | 64.4 | 69.1 |
| Agent Workflow | 41.3 | 61.7 | 80.2 |
| Code Generation | 42.6 | 53.1 | 60.4 |
| Role Play & Narrative | 70.5 | 73.2 | 76.4 |
| Research & Analysis | 63.9 | 69.4 | 77.3 |
Performance by hardware and tool
Every hardware/tool/quantization combination Ministral-3-8B-Instruct-2512-Q4_K_M has been benchmarked on, ranked by peak token generation speed. Last updated October 2026.
| Hardware | Tool | Quant | Best tok/s | Avg tok/s | Memory | Context | Quality | Runs |
|---|---|---|---|---|---|---|---|---|
| NVIDIA GeForce RTX 4060 Laptop GPU | llama.cpp | Q4_K_M | 35.4 tok/s | 34.9 tok/s | 7.7 GB | 32,768 tokens | 64.4 | 6 |
Benchmark runs
All 6 Ministral-3-8B-Instruct-2512-Q4_K_M runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is Ministral-3-8B-Instruct-2512-Q4_K_M good for coding?
- In our benchmarks, Ministral-3-8B-Instruct-2512-Q4_K_M scores 53.1/100 for coding. It runs at about 34.9 tok/s, so if you want more speed, nvidia/Qwen3.8-27B-NVFP4 is faster (~63.3 tok/s) and still scores well for coding (82.7/100).
- Is Ministral-3-8B-Instruct-2512-Q4_K_M good for agentic (tool-using) tasks?
- In our benchmarks, Ministral-3-8B-Instruct-2512-Q4_K_M scores 61.7/100 for agentic workflows. It runs at about 34.9 tok/s, so if you want more speed, Nex-N2.5-mini-APEX-Mini is faster (~88.2 tok/s) and still scores well for agentic workflows (91.3/100).
- How fast is Ministral-3-8B-Instruct-2512-Q4_K_M for local inference?
- Across 6 community benchmark runs, Ministral-3-8B-Instruct-2512-Q4_K_M reaches up to 35.4 tok/s and averages 34.9 tok/s, with the fastest results on NVIDIA GeForce RTX 4060 Laptop GPU.
- How much memory does Ministral-3-8B-Instruct-2512-Q4_K_M need?
- The leanest observed configuration used about 7.7 GB of memory (quantizations tested: Q4_K_M).
- Which tools have been used to run Ministral-3-8B-Instruct-2512-Q4_K_M?
- Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.