跳转到主要内容

Gemma 3 27B Instruct

GoogleGemmaOpen WeightGemma · Commercial OK

描述

Gemma 3 27B is a 27-billion-parameter vision-language model from Google, handling text and image input and generating text output. It features a 128K context window, multilingual support, and open weights. Suitable for complex question answering, summarization, reasoning, and image understanding tasks.

发布日期
2025-03-12
参数规模
27.0B
上下文长度
131K
支持模态
image, text

能力雷达图

26
general
12
coding
34
reasoning
28
science估算
0
agents
70
multimodal

Science 在缺少专门科学评测时使用推理能力代理估算。

排行榜排名

领域#排名分数来源
代码能力榜410
12.0
AA
通用能力榜396
26.0
AA
数学推理248
35.0
AA
多模态榜40
76.0
LS
推理能力79
54.0
LS
科学能力376
28.0
AA

基准测试分数 (LLM Stats)

Biology

GPQA42.4%自报

Code

HumanEval87.8%自报
LiveCodeBench29.7%自报

Factuality

FACTS Grounding74.9%自报
SimpleQA10.0%自报

Finance

MMLU-Pro67.5%自报

General

IFEval90.4%自报
Natural2Code84.5%自报
Global-MMLU-Lite75.1%自报
MBPP0.74 / 100自报
MMMU (val)64.9%自报
BIG-Bench Extra Hard19.3%自报

Image To Text

DocVQA86.6%自报
VQAv2 (val)71.0%自报
TextVQA65.1%自报

Language

BIG-Bench Hard87.6%自报
WMT24++53.4%自报
ECLeKTic16.7%自报

Math

GSM8k95.9%自报
MATH89.0%自报
MathVista-Mini67.6%自报
HiddenMath60.3%自报

Multimodal

AI2D84.5%自报
ChartQA78.0%自报
InfoVQA70.6%自报

Reasoning

Bird-SQL (dev)54.4%自报

AA 评测指数

Math Index
20.7
Intelligence Index
10.3
Coding Index
9.6
Math 500
0.9
Mmlu Pro
0.7
Gpqa
0.4
Ifbench
0.3
Aime
0.3
Scicode
0.2
Aime 25
0.2
Livecodebench
0.1
Tau2
0.1
Lcr
0.1
Hle
0.0
Terminalbench Hard
0.0

LLM Stats 分类评分

Structured Output
90
Instruction Following
90
Math
80
Vision
70
Finance
70
Grounding
70
Healthcare
70
Image To Text
70
Legal
70
Multimodal
70
Code
60
General
60
Language
60
Reasoning
60
Biology
40
Chemistry
40
Factuality
40
Physics
40

定价

输入价格$0.11 / 1M tokens
输出价格$0.25 / 1M tokens
混合价格(3:1)$0.145 / 1M tokens

速度

Tokens/秒31.4 tokens/s
首Token延迟0.73s
首回答延迟0.73s

可用提供商

(LS 内部计价单位)

暂无提供商数据

外部链接