Phi-4
MicrosoftPhiOpen WeightMIT · Commercial OK
描述
phi-4 is a state-of-the-art open model built to excel at advanced reasoning, coding, and knowledge tasks. It leverages a blend of synthetic data, filtered web data, academic texts, and supervised fine-tuning for precision, alignment, and safety.
發布日期
2024-12-12
參數規模
14.7B
上下文長度
16K
支援模態
text
能力雷達圖
28
general
17
coding
30
reasoning
36
science估算
0
agents
0
multimodal
Science 在缺少專門科學評測時使用推理能力代理估算。
排行榜排名
基準測試分數 (LLM Stats)
Biology
GPQA
56.1%自報
Code
HumanEval
82.6%自報
Creativity
Arena Hard
75.4%自報
Factuality
SimpleQA
3.0%自報
Finance
MMLU
84.8%自報
MMLU-Pro
70.4%自報
General
IFEval
63.0%自報
PhiBench
56.2%自報
LiveBench
47.6%自報
Math
MGSM
80.6%自報
MATH
80.4%自報
DROP
75.5%自報
Reasoning
HumanEval+
82.8%自報
AA 評測指數
Math Index18.0
Coding Index11.2
Intelligence Index10.4
Math 5000.8
Mmlu Pro0.7
Gpqa0.6
Scicode0.3
Ifbench0.2
Livecodebench0.2
Aime 250.2
Aime0.1
Hle0.0
Terminalbench Hard0.0
Lcr0.0
Tau20.0
LLM Stats 分類評分
Writing80
Code80
Creativity80
Finance80
Healthcare80
Language80
Legal80
Math70
Reasoning70
Structured Output60
Biology60
Chemistry60
General60
Instruction Following60
Physics60
Factuality0
定價
輸入價格$0.125 / 1M tokens
輸出價格$0.5 / 1M tokens
混合價格(3:1)$0.219 / 1M tokens
速度
Tokens/秒38.5 tokens/s
首Token延遲0.51s
首回答延遲0.51s
可用提供商
(LS 內部計價單位)暫無提供商資料