Phi-4: Complete Benchmark Performance Guide

other phi-4 2024-12-12 MIT

In-depth analysis of Phi-4 performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendorother
Versionphi-4
Release Date2024-12-12
Context Window16384 tokens
Input Modalitiestext
Output Modalitiestext
LicenseMIT
Documentationhttps://huggingface.co/models

Benchmark Performance

Benchmark Score Unit Evaluated At Notes Source
MMLU 69.2 % 2024-12-12 5-shot view
HUMANEVAL 62.8 pass@1 2024-12-12 — view
GSM8K 71.1 % 2024-12-12 0-shot CoT view
MATH 39.6 % 2024-12-12 0-shot CoT view
BBH 66.9 % 2024-12-12 3-shot CoT view
GPQA 30.1 % 2024-12-12 0-shot view
IFEVAL 59.6 % 2024-12-12 prompt_strict view
ARC 85.1 % 2024-12-12 challenge view
MUSR 38.6 % 2024-12-12 0-shot view
WINOGRANDE 78.3 % 2024-12-12 0-shot view

Pricing

Tier Price Currency
Input$0.15 / MtokUSD
Output$0.15 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://huggingface.co/models · as of 2024-12-12

Compliance

  • Data Residency: self-host
  • SOC2: ✗
  • HIPAA: ✗
  • GDPR: ✗
  • ISO 27001: ✗

Phi-4

Model Overview

Microsoft Phi-4 14B 模型, 16K 上下文, 改进推理与数学能力, 在 14B 规模上接近 GPT-4o-mini。

Core Specifications

Vendor Version Release Date Context Window Input Modalities Output Modalities License
Other phi-4 2024-12-12 16K text text MIT

Benchmark Performance

Benchmark Score Unit Notes
MMLU (Massive Multitask Language Understanding) 69.2 % 5-shot
HumanEval 62.8 pass@1 —
GSM8K (Grade School Math 8K) 71.1 % 0-shot CoT
MATH 39.6 % 0-shot CoT
BBH (BIG-Bench Hard) 66.9 % 3-shot CoT
GPQA 30.1 % 0-shot
IFEval 59.6 % prompt_strict
ARC 85.1 % challenge
MUSR 38.6 % 0-shot
WinoGrande 78.3 % 0-shot

Pricing

Input Output Cache Read Cache Write
— — — —

per million tokens

Strengths

  • Reliable general-purpose model.

Weaknesses

  • Proprietary, not self-hostable.
  • Context window 16K is limited.

Use Cases

  • General chat and Q&A

References