跳转到主要内容

Sarvam 30B (High)

Sarvam开源权重Apache 2.0 · 商用许可

描述

Sarvam-30B is an open-source 30B-parameter Mixture-of-Experts reasoning model from Sarvam AI trained from scratch and optimized for Indian languages, coding, and conversational workloads. It uses 128 sparse experts with 2.4B active parameters per token, Grouped Query Attention, and was pre-trained on 16 trillion tokens spanning code, mathematics, multilingual, and web data.

发布日期
2026-03-06
参数规模
30.0B
上下文长度
66K
支持模态
text

能力雷达图

70
general
60
coding
80
reasoning
60
science估算
40
agents
0
multimodal

缺少专门科学评测时,Science 由 LLM Stats 科学得分或推理能力估算。

排行榜排名

暂无排名数据

基准测试分数 (LLM Stats)

(LLM Stats (zeroeval))

General

MMLU85.1%自报
Arena-Hard v249.0%自报

Language

MMLU-Pro80.0%自报

Math

MATH-50097.0%自报
AIME 202596.7%自报
HMMT2574.2%自报
HMMT 202573.3%自报
Beyond AIME58.3%自报

Reasoning

MBPP0.93 / 100自报
HumanEvalOpenAI (2021)92.1%自报
LiveCodeBench v670.0%自报
GPQANYU + Cohere + Anthropic (2023)66.5%自报
BrowseCompOpenAI (2025)35.5%自报
SWE-Bench Verified34.0%自报

AA 评测指数

(Artificial Analysis)

暂无 AA 评测数据

LLM Stats 分类评分

(LLM Stats (zeroeval))
Language
80
Legal
80
Math
80
Finance
80
Healthcare
80
Physics
70
Reasoning
70
General
70
Biology
70
Chemistry
70
Code
60
Creativity
50
Writing
50
Search
40
Agents
40
Frontend Development
30

定价

暂无定价数据

速度

暂无速度数据

供应商价格排行

暂无提供商数据

外部链接