Model benchmarks
phi-3.1-mini-4k-instruct local LLM performance
As of October 2026, phi-3.1-mini-4k-instruct runs at up to 78.1 tok/s for local inference (best of 5 community benchmark runs across 2 GPUs).
Model size
Unknown
Peak speed
78.1 tok/s
Average speed
61.7 tok/s
Avg PP
1667.4 tok/s
Min memory
3.3 GB
Max context
32,768 tokens
Avg output / run
4,360 tokens
Avg runtime / run
4m 21s
Avg quality
25.8
Benchmark runs
5
GPUs tested
2
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 the 4 of 5 runs with completed quality scores. 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 | 23.2 | 25.8 | 29.9 |
| Agent Workflow | 32.0 | 38.4 | 45.8 |
| Code Generation | 0.8 | 3.7 | 6.9 |
| Role Play & Narrative | 26.8 | 38.3 | 52.5 |
| Research & Analysis | 7.3 | 20.1 | 31.0 |
Performance by hardware and tool
Every hardware/tool/quantization combination phi-3.1-mini-4k-instruct 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 3060 Laptop GPU | Bionic | — | 78.1 tok/s | 76.2 tok/s | 4.1 GB | 32,768 tokens | 26.7 | 4 |
| NVIDIA GeForce GTX 1050 | LM Studio | IQ3_M | 3.8 tok/s | 3.8 tok/s | 3.3 GB | 8,192 tokens | 23.3 | 1 |
Benchmark runs
All 5 phi-3.1-mini-4k-instruct runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is phi-3.1-mini-4k-instruct good for coding?
- In our benchmarks, phi-3.1-mini-4k-instruct scores 3.7/100 for coding. It runs at about 61.7 tok/s, so if you want more speed, Nex-N2.5-mini-IQ3_XXS is faster (~91.4 tok/s) and still scores well for coding (81.6/100).
- Is phi-3.1-mini-4k-instruct good for agentic (tool-using) tasks?
- In our benchmarks, phi-3.1-mini-4k-instruct scores 38.4/100 for agentic workflows. It runs at about 61.7 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 phi-3.1-mini-4k-instruct for local inference?
- Across 5 community benchmark runs, phi-3.1-mini-4k-instruct reaches up to 78.1 tok/s and averages 61.7 tok/s, with the fastest results on NVIDIA GeForce RTX 3060 Laptop GPU.
- How much memory does phi-3.1-mini-4k-instruct need?
- The leanest observed configuration used about 3.3 GB of memory (quantizations tested: IQ3_M).
- Which tools have been used to run phi-3.1-mini-4k-instruct?
- Benchmarks were submitted using Bionic, LM Studio. Results are community-contributed and updated as new runs arrive.