MiMo-V2.5-Pro

Xiaomi開源權重MIT · 商用許可

描述

MiMo-V2.5-Pro is Xiaomi's 1.02T-parameter sparse Mixture-of-Experts language model with 42B active parameters and a 1M-token context window. It inherits the MiMo-V2-Flash hybrid-attention and Multi-Token Prediction design, extends context during pre-training up to 1M tokens, and uses supervised fine-tuning, domain-specialized reinforcement learning, and Multi-Teacher On-Policy Distillation to improve complex software engineering, long-horizon agentic tasks, and ultra-long-context coherence.

發布日期

2026-04-22

參數規模

1.0T

上下文長度

1.0M

支援模態

audio, text

能力雷達圖

general

coding

reasoning

science

agents

multimodal

排行榜排名

領域	#排名	分數	來源
智慧體能力模型榜	55	48.0	LS
程式碼能力榜	134	76.0	AA
通用能力榜	81	69.0	AA
科學能力	103	73.0	AA

基準測試分數 (LLM Stats)

(LLM Stats (zeroeval))

Agents

TAU3-Bench

72.9%自報

WildClawBench

43.0%自報

Finance Agent v2

41.5%

Code

FrontierSWE (Impl.)

340.0%自報

MiMo Coding Bench

73.7%自報

Claw-Eval

64.0%自報

General

C-Eval

91.5%自報

MMLU

89.4%自報

Global-MMLU

83.6%自報

TriviaQA

81.3%自報

SWE-bench Verified (Agentless)

35.7%自報

Language

MMLU-Redux

92.8%自報

CMMLU

90.2%自報

MMLU-Pro

68.5%自報

Long Context

GraphWalks

62.0%自報

Math

GSM8k

99.6%自報

MATH

86.2%自報

AIMEMAA

37.3%自報

Reasoning

ARC-C

97.2%自報

HellaSwagAI2 (2019)

89.8%自報

BBH

88.4%自報

DROP

86.3%自報

Winogrande

85.6%自報

SWE-Bench Verified

78.9%自報

HumanEval+

75.6%自報

MBPP+

74.1%自報

Terminal-Bench 2.0Stanford × Laude Institute (2026)

68.4%自報

GPQANYU + Cohere + Anthropic (2023)

66.7%自報

SWE-Bench ProPrinceton NLP (2024)

57.2%自報

LiveCodeBench v6

39.6%自報

Humanity's Last Exam

34.0%自報

AA 評測指數

(Artificial Analysis)

Tau2(Sierra + U Toronto + Vector Institute (2025))

94.2

Gpqa(NYU + Cohere + Anthropic (2023))

86.6

Ifbench(Google Research (2023))

79.9

Lcr(Artificial Analysis)

79.7

Terminalbench V2 1

65.2

Coding Index(Artificial Analysis)

60.2

Scicode(UIUC + Argonne National Lab (2024))

50.6

Terminalbench Hard(Stanford × Laude Institute (2026))

43.2

Hle(Center for AI Safety + Scale AI (2025))

35.7

Intelligence Index(Artificial Analysis)

26.0

Tau Banking

9.9

Terminalbench V4 0

0.0

LLM Stats 分類評分

(LLM Stats (zeroeval))

Language

Legal

Math

Reasoning

Frontend Development

Healthcare

Physics

Finance

General

Biology

Chemistry

Code

Tool Calling

Long Context

Agents

Coding

Vision

定價

輸入價格$0.435 / 1M tokens

輸出價格$0.87 / 1M tokens

混合價格(3:1)$0.544 / 1M tokens

快取讀取價格$0.0036 / 1M tokens

速度

Tokens/秒25.5

首Token延遲1.92s

首回答延遲80.41s

供應商價格排行

4 個供應商

最便宜: Xiaomi最貴: EmpirioLabs AI

供應商輸入輸出

1Xiaomi最便宜

2DeepInfra

3Novita

$0.00001

4EmpirioLabs AI

$2.175

$4.35

比較該模型在不同 API 供應商之間的定價。

外部連結

LLM Stats Artificial Analysis