Microsoft
phi-3-mini 3.8B benchmark results
7 results on 4 benchmarks from 4 sources.
Open weights Released: Apr 23, 2024 Updated Oct 22, 2026
Also listed as: phi-3-mini-128k-instruct · phi-3-mini-4k-instruct · Phi-3-mini-4k-instruct · Phi-3-mini · microsoft/Phi-3-mini-4k-instruct · microsoft/Phi-3-mini-128k-instruct
All benchmark results
| Benchmarks | Score | Rank | Setting | Source type | Date |
|---|---|---|---|---|---|
| LMArena Text | 1130 | 387 | Overall, style control on; 20685 votes; 95% CI 1122–1137“phi-3-mini-128k-instruct” | Official leaderboard | Oct 8, 2026 |
| LMArena Text | 1128 | 388 | Overall, style control on; 20118 votes; 95% CI 1122–1135“phi-3-mini-4k-instruct” | Official leaderboard | Oct 8, 2026 |
| Open LLM Leaderboard v2 | 27.6% | 1339 | Average of 6 normalised benchmarks; float16; 3.821B params“microsoft/Phi-3-mini-4k-instruct” | Official leaderboard | Mar 13, 2025 |
| Open LLM Leaderboard v2 | 26.3% | 1459 | Average of 6 normalised benchmarks; bfloat16; 3.821B params“microsoft/Phi-3-mini-128k-instruct” | Official leaderboard | Mar 13, 2025 |
| MMLU | 68.8% | 28 | Exact match; 5-shot | Paper / tech report via Epoch AI | Apr 23, 2024 |
| HumanEval / HumanEval+ | 64.6% | 52 | HumanEval (base tests), pass@1 greedy“Phi-3-mini-4k-instruct” | Official leaderboard | Dec 26, 2024 |
| HumanEval / HumanEval+ | 59.1% | 52 | HumanEval+ (extended tests), pass@1 greedy“Phi-3-mini-4k-instruct” | Official leaderboard | Dec 26, 2024 |
FAQ
Where do these phi-3-mini 3.8B scores come from?
From EvalPlus leaderboard, LMArena leaderboard dataset, Epoch AI Benchmarking Hub, and Hugging Face Open LLM Leaderboard v2 (archived). Each row links to the original leaderboard, dataset or report, and vendor-reported numbers are labelled.