MiMo-V2.5-Pro
XiaomiOpen WeightMIT · Usage Commercial
Description
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.
Date de sortie
2026-04-22
Paramètres
1.0T
Longueur du contexte
1.0M
Modalités
text
Radar de capacités
41
general
59
coding
87
reasoning
64
science
70
agents
0
multimodal
Classements
| Domaine | #Rang | Score | Source |
|---|---|---|---|
| Capacité agentique | 52 | 52.0 | LS |
| Classement codage | 64 | 78.0 | AA |
| Classement général | 41 | 81.0 | AA |
| Science | 56 | 79.0 | AA |
Scores de benchmarks (LLM Stats)
(LLM Stats (zeroeval))Agents
FrontierSWE (Impl.)
340.0%Aut.
MiMo Coding Bench
73.7%Aut.
TAU3-Bench
72.9%Aut.
Terminal-Bench 2.0Stanford × Laude Institute (2026)
68.4%Aut.
Claw-Eval
64.0%Aut.
SWE-Bench ProPrinceton NLP (2024)
57.2%Aut.
WildClawBench
43.0%Aut.
Finance Agent v2
41.5%
Biology
GPQANYU + Cohere + Anthropic (2023)
66.7%Aut.
Code
SWE-Bench Verified
78.9%Aut.
Finance
MMLU
89.4%Aut.
MMLU-Pro
68.5%Aut.
General
ARC-C
97.2%Aut.
MMLU-Redux
92.8%Aut.
C-Eval
91.5%Aut.
CMMLU
90.2%Aut.
Global-MMLU
83.6%Aut.
TriviaQA
81.3%Aut.
MBPP+
74.1%Aut.
LiveCodeBench v6
39.6%Aut.
SWE-bench Verified (Agentless)
35.7%Aut.
Language
BBH
88.4%Aut.
Winogrande
85.6%Aut.
Long Context
GraphWalks
62.0%Aut.
Math
GSM8k
99.6%Aut.
DROP
86.3%Aut.
MATH
86.2%Aut.
AIMEMAA
37.3%Aut.
Humanity's Last Exam
34.0%Aut.
Reasoning
HellaSwagAI2 (2019)
89.8%Aut.
HumanEval+
75.6%Aut.
Indices d'évaluation AA
(Artificial Analysis)Coding Index(Artificial Analysis)60.2
Intelligence Index(Artificial Analysis)42.9
Tau2(Sierra + U Toronto + Vector Institute (2025))0.9
Gpqa(NYU + Cohere + Anthropic (2023))0.9
Ifbench(Google Research (2023))0.8
Lcr(Artificial Analysis)0.8
Terminalbench V2 10.7
Scicode(UIUC + Argonne National Lab (2024))0.5
Terminalbench Hard(Stanford × Laude Institute (2026))0.4
Hle(Center for AI Safety + Scale AI (2025))0.4
Tau Banking0.1
Scores par catégorie 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
Tarification
Prix d'entrée$0.435 / 1M tokens
Prix de sortie$0.87 / 1M tokens
Prix mixte (3:1)$0.544 / 1M tokens
Prix de lecture cache$0.0036 / 1M tokens
Vitesse
Tokens/sec48.1
Délai du premier token2.82s
Temps de réponse44.40s
Classement des Prix par Fournisseur
Classement des Prix par Fournisseur
8 fournisseurs
Moins cher: XiaomiPlus cher: EmpirioLabs AI
FournisseurEntréeSortie
1XiaomiMoins cher
$0
$0
2DeepInfra
$0
$0
3Novita
$0
$0.00001
4DigitalOcean
$0.4
$1.5
5CrofAI
$0.4
$0.8
6OpenCode Go
$0.435
$0.87
7LLM Gateway
$0.435
$0.87
8EmpirioLabs AI
$2.175
$4.35
Comparer les prix entre différents fournisseurs API pour ce modèle.