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
18
general
7
coding
18
reasoning
24
science
15
agents
0
multimodal
排行榜排名
基准测试分数 (LLM Stats)
(LLM Stats (zeroeval))General
MMLU
67.3%自报
TruthfulQA
66.4%自报
Multilingual MMLU
49.3%自报
Arena Hard
32.8%自报
Language
BoolQ
81.2%自报
MMLU-Pro
52.8%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
ARC-C
83.7%自报
OpenBookQA
79.2%自报
PIQA
77.6%自报
Social IQa
72.5%自报
BIG-Bench Hard
70.4%自报
HellaSwagAI2 (2019)
69.1%自报
Winogrande
67.0%自报
GPQANYU + Cohere + Anthropic (2023)
52.0%自报
AA 评测指数
(Artificial Analysis)Math 500(OpenAI (2024), subset of Hendrycks et al. MATH (2021))69.6
Mmlu Pro(TIGER-Lab (Univ. of Waterloo, Toronto, CMU, 2024))46.5
Gpqa(NYU + Cohere + Anthropic (2023))33.1
Ifbench(Google Research (2023))21.1
Lcr(Artificial Analysis)15.3
Livecodebench(UC Berkeley + MIT + Cornell (2024))12.6
Tau2(Sierra + U Toronto + Vector Institute (2025))8.2
Math Index(Artificial Analysis)6.7
Aime 25(MAA (Mathematical Association of America))6.7
Intelligence Index(Artificial Analysis)6.3
Hle(Center for AI Safety + Scale AI (2025))4.4
Coding Index(Artificial Analysis)3.8
Aime(MAA (Mathematical Association of America))3.0
Terminalbench V2 10.4
Terminalbench Hard(Stanford × Laude Institute (2026))0.0
LLM Stats 分类评分
(LLM Stats (zeroeval))Math70
Psychology70
Reasoning70
General70
Language60
Legal60
Finance60
Healthcare60
Physics50
Creativity50
Chat30
Biology30
Chemistry30
Writing30
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46.0
首Token延迟0.33s
首回答延迟0.33s
供应商价格排行
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。