Ministral 3 8B
MistralMistral开源权重Mistral Research License
描述
The Ministral-8B-Instruct-2410 is an instruct fine-tuned model for local intelligence, on-device computing, and at-the-edge use cases, significantly outperforming existing models of similar size.
发布日期
2025-12-02
参数规模
8.0B
上下文长度
—
支持模态
image, text
能力雷达图
23
general
18
coding
34
reasoning
29
science
29
agents
80
multimodal
排行榜排名
基准测试分数 (LLM Stats)
(LLM Stats (zeroeval))Chat
MT-Bench
0.83 / 100自报
MM-MT-Bench
0.08 / 100自报
General
Arena Hard
70.9%自报
Multilingual MMLU
70.6%自报
Wild Bench
66.8%自报
TriviaQA
65.5%自报
MMLU
65.0%自报
French MMLU
57.5%自报
Language
MMLU-Redux
79.3%自报
Math
AIME 2024
86.0%自报
AIME 2025
78.7%自报
MATH (CoT)
62.6%自报
MATH
54.5%自报
Reasoning
Winogrande
75.3%自报
ARC-C
71.9%自报
MBPP pass@1
70.0%自报
GPQANYU + Cohere + Anthropic (2023)
66.8%自报
LiveCodeBench
61.6%自报
AGIEval
48.3%自报
HumanEvalOpenAI (2021)
34.8%自报
AA 评测指数
(Artificial Analysis)Mmlu Pro(TIGER-Lab (Univ. of Waterloo, Toronto, CMU, 2024))64.2
Gpqa(NYU + Cohere + Anthropic (2023))47.1
Math Index(Artificial Analysis)31.7
Aime 25(MAA (Mathematical Association of America))31.7
Livecodebench(UC Berkeley + MIT + Cornell (2024))30.3
Ifbench(Google Research (2023))29.1
Tau2(Sierra + U Toronto + Vector Institute (2025))26.6
Lcr(Artificial Analysis)25.7
Scicode(UIUC + Argonne National Lab (2024))20.7
Coding Index(Artificial Analysis)9.7
Intelligence Index(Artificial Analysis)5.5
Terminalbench Hard(Stanford × Laude Institute (2026))4.5
Hle(Center for AI Safety + Scale AI (2025))4.3
Terminalbench V2 14.1
Tau Banking3.7
LLM Stats 分类评分
(LLM Stats (zeroeval))Chat80
Roleplay80
Communication80
Creativity80
Language70
Finance70
Healthcare70
Writing70
Legal60
Math60
Reasoning60
General60
Code30
定价
输入价格$0.15 / 1M tokens
输出价格$0.15 / 1M tokens
混合价格(3:1)$0.15 / 1M tokens
速度
Tokens/秒104.6
首Token延迟0.46s
首回答延迟0.46s
供应商价格排行
供应商价格排行
4 个供应商
最便宜: Mistral最贵: Kilo Gateway
供应商输入输出
1Mistral最便宜
$0.1
$0.1
2NanoGPT
$0.15
$0.15
3OpenRouter
$0.15
$0.15
4Kilo Gateway
$0.15
$0.15
比较该模型在不同 API 供应商之间的定价。