BestLLMfor EN Your hardware. Your LLM. Your call.
APIOpen data Find my LLM
Model fiche

Phi-3.5 Mini

By Microsoft · United States

Updated 2026-07-13

chat small
Parameters
3.8B
License
MIT
Context
128k
VRAM (Q4)
10 GB
Released
August 2024

Overview

Microsoft's Phi-3.5 Mini, a 3.8B model trained on heavily curated synthetic data with a 128k context. Punches above its weight on reasoning.

When to pick this model

  • Long-context tasks on hardware that can't fit a 7B
  • Reasoning-heavy workloads at small size
  • MIT-licensed embedded or commercial deployments
  • Latency-critical assistants on consumer hardware
  • STEM-focused tutoring and Q&A apps

VRAM requirements by quantization

VRAM REQUIRED (GB)812162432Q4_K_M10 GBQ5_K_M12 GBQ8_018 GBFP1633 GB
QuantizationVRAM required
Q4_K_M (recommended)10 GB
Q5_K_M12 GB
Q8_018 GB
FP16 (no quantization)33 GB

VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.

In practice, Phi-3.5 Mini needs a 12 GB card at Q4_K_M (10 GB). Stepping up to Q8_0 nearly doubles the footprint to 18 GB, and unquantized FP16 weights take 33 GB — plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity.

Without a GPU, Phi-3.5 Mini needs roughly 12 GB of system RAM to run on CPU via llama.cpp or Ollama — workable for background jobs, but far slower than GPU inference. Throughput estimates from our compatibility engine: around 14 tokens/sec on entry-level GPUs, on the order of 40 tokens/sec on a mid-range card, and up to 100 tokens/sec on high-end hardware — assuming the chosen quantization fully fits in VRAM.

What hardware do you need

The table below matches Phi-3.5 Mini to common GPU memory tiers, using the highest-fidelity quantization that fully fits each card class. Spilling layers to system RAM works but costs most of the speed, so size your card to the quantization you actually want to run.

GPU memoryExample cardsBest fit for Phi-3.5 Mini
8 GBRTX 5070 Laptop, RTX 5060, RTX 5060 Ti 8GBDoes not fit — needs 10 GB at Q4_K_M
12 GBRTX 5070, RTX 5070 Ti Laptop, RTX 4080 LaptopQ5_K_M (12 GB used)
16 GBRTX 5080, RTX 4080 Super, Radeon RX 9070 XTQ5_K_M (12 GB used)
24 GBRTX 4090, Radeon RX 7900 XTX, RTX 5090 LaptopQ8_0 (18 GB used)
32 GBRTX 5090Q8_0 (18 GB used)

Which GPU should you buy to run Phi-3.5 Mini?

To run Phi-3.5 Mini locally at Q4, you need ~10 GB of VRAM. The best value for this is a RTX 5070 (12 GB VRAM).

Check RTX 5070 price on Amazon →

As an Amazon Associate, BestLLMfor earns from qualifying purchases, at no extra cost to you. It does not influence our independent rankings.

Published benchmark scores

BenchmarkScore
MMLU69
HumanEval62.8

Scores published by the model author or aggregated from public leaderboards. Re-measured monthly by our editorial team.

To put Phi-3.5 Mini in context: its MMLU score of 69 ranks #23 of the 34 catalog models with a published MMLU result (catalog median 73.4); its HumanEval score of 62.8 ranks #22 of the 26 catalog models with a published HumanEval result (catalog median 81.1). Scores are author-published and measured under different harnesses, so treat gaps of a few points as noise rather than a verdict.

Strengths

  • 128k context in a 3.8B footprint
  • MIT license with no commercial restrictions
  • Fast inference on modest hardware
  • Strong reasoning relative to its size

Limitations

  • Memory footprint is high for a 3.8B at full context
  • Outclassed by Phi-4 14B on overall quality
  • Synthetic-heavy training can show as a narrow knowledge base

Typical workloads

In our catalog grid, Phi-3.5 Mini is filed under Lightweight RAG, Laptop — the use cases where its size/quality trade-off makes the most sense. Its tags translate to concrete workloads: everyday chat, drafting and summarization.

The 128k-token context window is large enough to hold entire codebases' worth of files or long reports in a single prompt, which is what makes local RAG and document analysis practical. The MIT license is permissive, so shipping it inside a commercial product raises no special legal questions.

Architecture & training

Architecture: Dense · 3.8B · Phi-3.5 Mini · sliding window + FlashAttention

Training: High-quality synthetic data from Microsoft. Strong educational focus.

Verdict

A clever small model with a huge context — useful when you need 128k tokens and minimal VRAM.

Quick start

ollama run phi3.5

Or use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.

Similar models worth comparing

Frequently asked questions

How much VRAM does Phi-3.5 Mini need?

At the recommended Q4_K_M quantization, Phi-3.5 Mini needs about 10 GB of VRAM. Q8_0 takes 18 GB, and unquantized FP16 weights take 33 GB.

Can Phi-3.5 Mini run without a GPU?

Yes — with roughly 12 GB of system RAM it runs CPU-only through llama.cpp or Ollama. Expect a fraction of GPU speed, which is fine for background or batch jobs but slow for interactive chat.

What context window does Phi-3.5 Mini support?

Phi-3.5 Mini supports a 128k-token context window (131,072 tokens).

Can I use Phi-3.5 Mini commercially?

Yes. Phi-3.5 Mini is released under MIT, a permissive open-source license that allows commercial use, modification and redistribution.

How fast is Phi-3.5 Mini on consumer hardware?

Our compatibility engine estimates on the order of 40 tokens/sec on a mid-range GPU and up to 100 tokens/sec on high-end cards, assuming the quantization fully fits in VRAM.

Which quantization of Phi-3.5 Mini should I download first?

Start with Q4_K_M (10 GB) — the standard size/quality sweet spot. Step up to Q5_K_M or Q8_0 only if you have VRAM headroom. On a 24 GB card you can run up to Q8_0.

Tools

Is Phi-3.5 Mini the right pick for you?

Compute self-hosted ROI → Back to catalog