MiMo-V2.5-Pro
XiaomiOpen WeightMIT · Uso Comercial
Descripción
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.
Fecha de lanzamiento
2026-04-22
Parámetros
1.0T
Longitud del contexto
1.0M
Modalidades
audio, text
Radar de capacidades
28
general
59
coding
87
reasoning
64
science
70
agents
30
multimodal
Rankings
| Dominio | #Posición | Puntuación | Fuente |
|---|---|---|---|
| Capacidad agéntica | 55 | 48.0 | LS |
| Ranking de codificación | 134 | 76.0 | AA |
| Ranking general | 81 | 69.0 | AA |
| Ciencia | 103 | 73.0 | AA |
Puntuaciones de benchmarks (LLM Stats)
(LLM Stats (zeroeval))Agents
TAU3-Bench
72.9%Aut.
WildClawBench
43.0%Aut.
Finance Agent v2
41.5%
Code
FrontierSWE (Impl.)
340.0%Aut.
MiMo Coding Bench
73.7%Aut.
Claw-Eval
64.0%Aut.
General
C-Eval
91.5%Aut.
MMLU
89.4%Aut.
Global-MMLU
83.6%Aut.
TriviaQA
81.3%Aut.
SWE-bench Verified (Agentless)
35.7%Aut.
Language
MMLU-Redux
92.8%Aut.
CMMLU
90.2%Aut.
MMLU-Pro
68.5%Aut.
Long Context
GraphWalks
62.0%Aut.
Math
GSM8k
99.6%Aut.
MATH
86.2%Aut.
AIMEMAA
37.3%Aut.
Reasoning
ARC-C
97.2%Aut.
HellaSwagAI2 (2019)
89.8%Aut.
BBH
88.4%Aut.
DROP
86.3%Aut.
Winogrande
85.6%Aut.
SWE-Bench Verified
78.9%Aut.
HumanEval+
75.6%Aut.
MBPP+
74.1%Aut.
Terminal-Bench 2.0Stanford × Laude Institute (2026)
68.4%Aut.
GPQANYU + Cohere + Anthropic (2023)
66.7%Aut.
SWE-Bench ProPrinceton NLP (2024)
57.2%Aut.
LiveCodeBench v6
39.6%Aut.
Humanity's Last Exam
34.0%Aut.
Índices de evaluación AA
(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 165.2
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 Banking9.9
Terminalbench V4 00.0
Puntuaciones por categoría LLM Stats
(LLM Stats (zeroeval))Language90
Legal80
Math80
Reasoning80
Frontend Development80
Healthcare80
Physics70
Finance70
General70
Biology70
Chemistry70
Code70
Tool Calling70
Long Context60
Agents60
Coding40
Vision30
Precios
Precio de entrada$0.435 / 1M tokens
Precio de salida$0.87 / 1M tokens
Precio mixto (3:1)$0.544 / 1M tokens
Precio de lectura caché$0.0036 / 1M tokens
Velocidad
Tokens/seg25.5
Retraso del primer token1.92s
Tiempo hasta la respuesta80.41s
Ranking de Precios por Proveedor
Ranking de Precios por Proveedor
4 proveedores
Más barato: XiaomiMás caro: EmpirioLabs AI
ProveedorEntradaSalida
1XiaomiMás barato
$0
$0
2DeepInfra
$0
$0
3Novita
$0
$0.00001
4EmpirioLabs AI
$2.175
$4.35
Comparar precios entre diferentes proveedores de API para este modelo.