Inference
EN
Xiaomi

MiMo-V2.5-Pro

MiMo-V2.5-Pro is a reasoning model from Xiaomi in the MiMo-V2.5 family. 27 benchmarks count toward its index, in 6 categories.

  • Open weights
  • Reasoning
  • Input: text
  • Released 2026-04
RankedEnough results for an index.
IndexAll categories, out of 100.56.3±4.6Median 51.0
CoverageIndex weight that has results.85%
SpeedOutput tokens per second.47/sMedian 67/s
First tokenSeconds to the first output token.1.6 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.48Median $0.60
Output / 1MUS dollars per 1M output tokens.$1.80Median $2.55
ContextMaximum tokens in one request.1.02MMedian 262K
EloLMArena rating and rank.1464±4 #38Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
xiaomi/mimo-v2.5-pro
Max outputMaximum output tokens per request.
236K
Knowledge cutoffLast date of training data.
—
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
Yes
Cache read / 1MUS dollars per 1M cached tokens.
$0.096
WeightsLink to the published weights.
Hugging Face (opens in a new tab) ↗
RetiresDate OpenRouter removes the model.
—

Where to run

APIs that serve this model.
5 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
XiaomiDeprecatedMaker$0.435$0.871.05M
Novita AIHost$0.522$1.041.05M
OpenRouterDeprecatedRouter$0.48$1.801.02M
DigitalOceanCloud$0.80$3.00262K
Deep InfraDeprecatedHost$1.00$3.001.05M

US dollars per 1M tokens, list price. From models.dev (opens in a new tab) (MIT).

Run locally

1023.18B parameters
16-bit
2149.7 GB
8-bit
1142.5 GB
4-bit
652.3 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy needs more than 128 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.13 counted58.281 / 218
CodingCode writing and repair.6 counted57.283 / 209
ReasoningLogic problems and puzzles.2 counted53.395 / 227
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.4 counted51.2125 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.1 counted63.7—
MathMath problems.1 counted54.8—

Results

27 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
Artificial Analysis IFBenchInstruction79.9%5 / 146
τ³-Bench Tool-Agent-User EvaluationAgentic72.9%3 / 32
Gert Labs Composite Game BenchmarkAgentic62.7%9 / 52
τ²-Bench Tool-Agent-User EvaluationAgentic94.2%26 / 149
Terminal-Bench 2.0 · BenchLMAgentic68.4%8 / 41
Artificial Analysis Long Context ReasoningReasoning79.7%45 / 217
Artificial Analysis Humanity's Last ExamKnowledge35.7%59 / 222
Humanity's Last ExamKnowledge48.0%23 / 77
Artificial Analysis GPQA DiamondKnowledge86.6%63 / 199
Artificial Analysis SciCodeCoding50.6%49 / 114
Critical Physics TasksReasoning4.0%92 / 214
LiveCodeBenchCoding81.4% ±1.162 / 142
GPQA Diamond · Vals AIKnowledge82.6% ±2.559 / 135
MMLU ProKnowledge84.6% ±0.560 / 135
Claw-EvalAgentic63.8%19 / 42
GDPval-AA normalizedAgentic31.0%61 / 133
MazeBench: visual spatial reasoning in a 3D open worldAgentic0.0%23 / 48
SWE-bench ProCoding57.2%44 / 83
SWE-benchCoding74.0% ±2.050 / 87
Terminal-Bench 2.1 · Vals AIAgentic57.3% ±5.344 / 75
Humanity's Last Exam without toolsKnowledge34.0%34 / 53
Vibe Code Bench v1.1Coding34.1% ±4.579 / 123
Code MigrationCoding21.6% ±4.150 / 77
Artificial Analysis Omniscience AccuracyKnowledge22.4%119 / 164
ProofBench v1.1Math22.0% ±4.242 / 54
Artificial Analysis ITBench-AAAgentic38.2%20 / 24
Agent Arena command recoveryAgentic−7.545 / 50
Terminal-Bench HardAgentic43.2%21 / 23
Terminal-Bench 4.0Agentic0.5% ±0.557 / 61
APEX-Agents-AAAgentic2.4%29 / 30
TaxEval v2Finance73.8% ±0.946 / 143
MedScribeHealthcare83.7% ±2.142 / 109
CorpFin v2Finance61.4% ±1.052 / 131
EMBFinance55.2% ±3.149 / 75
LegalBenchLegal77.1% ±0.6111 / 149
Finance Agent (v2)Finance41.5% ±1.261 / 79
Legal Research BenchLegal15.9% ±2.570 / 90
Tax Agent BenchFinance11.5% ±2.066 / 82
MedCodeHealthcare32.5% ±1.988 / 106

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseVals AINo licence statedLMArenaCC BY 4.0models.devMIT

History

5 releases
Index57.2 → 56.3
23 Oct 202612 Nov 2026
Output / 1M$0.87 → $1.80
23 Oct 202612 Nov 2026

Questions

How much does MiMo-V2.5-Pro cost?
$0.48 per 1M input tokens, $1.80 per 1M output tokens.
How good is MiMo-V2.5-Pro?
Its index is 56.3 out of 100. That is place 93 of 283 ranked models.
What is MiMo-V2.5-Pro best at?
Agentic: 58.2 out of 100, place 81 of 218.
How fast is MiMo-V2.5-Pro?
47 output tokens per second, 1.6 s to the first token.
What is the context window of MiMo-V2.5-Pro?
1.02M tokens.
Is MiMo-V2.5-Pro open source?
Yes. Its weights are published.
When was MiMo-V2.5-Pro released?
April 2026.

Badge

MiMo-V2.5-Pro on Inference360: place 93 of 283, index 56.3
HTML
<a href="https://inference360.ai/models/mimo-v2-5-pro"><img src="https://inference360.ai/badges/mimo-v2-5-pro.svg" alt="MiMo-V2.5-Pro on Inference360: place 93 of 283, index 56.3" height="20"></a>
Markdown
[![MiMo-V2.5-Pro on Inference360: place 93 of 283, index 56.3](https://inference360.ai/badges/mimo-v2-5-pro.svg)](https://inference360.ai/models/mimo-v2-5-pro)

Compare with

Claude Opus 4.5 Thinking56.3GLM-5.156.0Claude Sonnet 4.657.0Qwen 3.6 Max (preview)57.2MiMo-V2.554.8

Lower price

dots3-note Preview63.4 · FreeApodex 1.1 Mini55.7 · FreeGLM-5.3-Flash63.1 · $0.078/1M

More from Xiaomi

All models
MiMo-V2.554.8MiMo-V2-Pro51.8MiMo-V2-Omni47.2MiMo-V2-Flash43.4MiMo V2.6 Pro UltraSpeedUnranked