AI Potluck
Model components / Fine-tuned / chat models

Phi-4-mini-instruct

Microsoft

Compact instruction-tuned small language model (3.8B) that scores competitively with larger models on coding benchmarks. Part of Microsoft's Phi-4 family, which expanded across 2025–early 2026 with Phi-4-reasoning, Phi-4-multimodal, and Phi-4-Reasoning-Vision-15B (March 2026); no Phi-5 has shipped as of May 2026.

Phi-4-mini-instruct, 3.8B dense, 128k context, 200k vocab, GQA; released Feb 2025 under MIT. SFT/DPO-aligned small instruct model. Verified 2026-08-13 via the HF model card.

Openness

3 high confidence
3.0
weights
open(MIT, on HF)
data
closed(synthetic-heavy training data undisclosed)
code
closed(no training pipeline)
license
MIT(OSI permissive)

MIT-licensed open weights, with the training and post-training data and the training code unreleased, so this is open weights rather than open source.

  • https://huggingface.co/microsoft/Phi-4-mini-instruct recorded 2026-08-13

    `license:mit` in the repo tags and `"gated":false` in the embedded repo state, so the 3.8B weights download without a barrier; the card describes the data mixture in prose and publishes neither the corpus nor any training or post-training code

Adoption

3 high confidence
3.0

The one declared artifact, microsoft/Phi-4-mini-instruct, reads 429,490 downloads in the trailing 30 days, which bands at level 3 (100K-1M) on the software and model adoption scale.

Capability

2 high confidence
2.0

Strong-for-size 3.8B with good math (GSM8K 88.6) but low GPQA (25.2) and overall well below 2026 frontier chat models; small-model tier on a frontier-anchored scale.

  • https://huggingface.co/microsoft/Phi-4-mini-instruct recorded 2026-08-13

    quality table, Phi-4 mini-Ins column - MMLU (5-shot) 67.3, MMLU-Pro (0-shot, CoT) 52.8, GPQA (0-shot, CoT) 25.2, GSM8K (8-shot, CoT) 88.6, MATH (0-shot, CoT) 64.0, Overall 63.5; GPT-4o-mini column Overall 75.5

Verified 2026-08-13