Phi-4 Mini Instruct

MicrosoftPhi开源权重MIT · 商用许可

描述

Phi 4 Mini Instruct is a lightweight (3.8B parameters) open model built upon synthetic data and filtered web data, focusing on high-quality reasoning. It supports a 128K token context length and is enhanced for instruction adherence and safety via supervised fine-tuning and direct preference optimization.

发布日期

2024-02-26

参数规模

3.8B

上下文长度

—

支持模态

—

能力雷达图

general

coding

reasoning

science

agents

multimodal

排行榜排名

领域	#排名	分数	来源
代码能力榜	515	9.0	AA
通用能力榜	526	17.0	AA
科学能力	518	17.0	AA

基准测试分数 (LLM Stats)

(LLM Stats (zeroeval))

Biology

GPQANYU + Cohere + Anthropic (2023)

52.0%自报

Creativity

Social IQa

72.5%自报

Arena Hard

32.8%自报

Finance

MMLU

67.3%自报

TruthfulQA

66.4%自报

MMLU-Pro

52.8%自报

General

ARC-C

83.7%自报

OpenBookQA

79.2%自报

PIQA

77.6%自报

Multilingual MMLU

49.3%自报

Language

BoolQ

81.2%自报

BIG-Bench Hard

70.4%自报

Winogrande

67.0%自报

Math

MATH-500

94.6%自报

GSM8k

88.6%自报

MATH

64.0%自报

MGSM

63.9%自报

AIMEMAA

57.5%自报

Reasoning

HellaSwagAI2 (2019)

69.1%自报

AA 评测指数

(Artificial Analysis)

Math Index(Artificial Analysis)

6.7

Intelligence Index(Artificial Analysis)

5.7

Coding Index(Artificial Analysis)

3.8

Math 500(OpenAI (2024), subset of Hendrycks et al. MATH (2021))

0.7

Mmlu Pro(TIGER-Lab (Univ. of Waterloo, Toronto, CMU, 2024))

0.5

Gpqa(NYU + Cohere + Anthropic (2023))

0.3

Ifbench(Google Research (2023))

0.2

Lcr(Artificial Analysis)

0.2

Livecodebench(UC Berkeley + MIT + Cornell (2024))

0.1

Scicode(UIUC + Argonne National Lab (2024))

0.1

Tau2(Sierra + U Toronto + Vector Institute (2025))

0.1

Aime 25(MAA (Mathematical Association of America))

0.1

Hle(Center for AI Safety + Scale AI (2025))

0.0

Aime(MAA (Mathematical Association of America))

0.0

Terminalbench V2 1

0.0

Terminalbench Hard(Stanford × Laude Institute (2026))

0.0

LLM Stats 分类评分

(LLM Stats (zeroeval))

Math

Psychology

Reasoning

General

Legal

Language

Finance

Healthcare

Physics

Creativity

Biology

Chemistry

Writing

定价

输入价格免费

输出价格免费

混合价格(3:1)免费

速度

Tokens/秒45.1

首Token延迟0.34s

首回答延迟0.34s

供应商价格排行

2 个供应商

最便宜: Azure Cognitive Services最贵: Azure

供应商输入输出

1Azure Cognitive Services最便宜

$0.075

$0.3

2Azure

$0.075

$0.3

比较该模型在不同 API 供应商之间的定价。

外部链接

LLM Stats Artificial Analysis