LLM
MiMo-V2-Flash
Xiaomi MoE language model built for fast reasoning, coding and agentic workflows, with a hybrid thinking mode and a 256K context window.
Levelmodel
ClassLLM
Parameters309B
Context256,000
Released16 Dec 2025
Opennessopen weights permissive
Runs onnot in the notes
Statusactive
Benchmark Results
vendor-reportedindependent
AIME 2025 2025math
vendor-reported
94.1
050100
Score in %. Higher is better.
MiMo-V2-Flash model card, post-trained model table
Artificial Analysis Intelligence Index v4.3.2composite index
Intelligence Index v4.3.2; marked estimated (asterisk) on page; AA entry MiMo-V2-Flash (Feb 2026)independent
22
03060
Score in index. Higher is better.
rank 40 of 117, AA model page, rank within model class
BrowseCompagents and tool use
Without context managementvendor-reported
45.4
050100
Score in %. Higher is better.
MiMo-V2-Flash model card, post-trained model table
GPQA Diamondknowledge and reasoning
vendor-reported
83.7
050100
Score in %. Higher is better.
MiMo-V2-Flash model card, post-trained model table
Humanity's Last Exam (HLE)knowledge and reasoning
No toolsvendor-reported
22.1
050100
Score in %. Higher is better.
MiMo-V2-Flash model card, post-trained model table
LiveCodeBench v6coding
vendor-reported
80.6
050100
Score in %. Higher is better.
MiMo-V2-Flash model card, post-trained model table
MMLU-Proknowledge and reasoning
vendor-reported
84.9
050100
Score in %. Higher is better.
MiMo-V2-Flash model card, post-trained model table
SWE-bench Multilingualcoding
vendor-reported
71.7
050100
Score in %. Higher is better.
MiMo-V2-Flash model card, post-trained model table
SWE-bench Verifiedcoding
vendor-reported
73.4
050100
Score in %. Higher is better.
MiMo-V2-Flash model card, post-trained model table
Terminal-Bench 2.0agents and tool use
vendor-reported
38.5
050100
Score in %. Higher is better.
MiMo-V2-Flash model card, post-trained model table
Where this comes from
Every score names who ran it: vendor-reported, maintainer-run or independent. A rank is shown with the population it was ranked in (gap rule 3).
Lineage
continued pretrain
MiMo-V2.5 is a child (continued pretrain)high confidencemodel cardWhere this comes from
Parent and child relations from model cards, Hugging Face base_model fields and the genealogy tables of the research notes. The tree shows two generations each way, 2 models. Confidence and evidence are the notes' own.
R&D Topics
Identity
Where this comes from
Developer and dates from the research notes. Release date is the vendor's public announcement; weights date is when open weights were published.
Architecture and Size
- Architecture
- MoE Hybrid attention interleaving sliding-window (128-token window) and global attention at 5:1, with multi-token prediction
- Total parameters
- 309B sources
- Active parameters
- 15B
- Checkpoint parameters
- –
- Context length
- 256,000 tokens
- Input
- text
- Output
- text
Where this comes from
Parameter counts as the developer states them. Where the checkpoint holds a different total, both are shown (register A16).
Openness and Access
- Openness
- open weights permissive
- Licence
- MIT License sources
- Access routes
- API, open weights download
Where this comes from
The licence file governs over registry tags and aggregator labels.
Deployment
- Runs on
- not in the notes
- Footprint
- Trained on 27T tokens with FP8 mixed precision. Artificial Analysis tracks a February 2026 version as MiMo-V2-Flash (Feb 2026)
- Runtimes
- not in the notes
Where this comes from
Deployment tiers and runtimes as named in the notes. Footprint is the notes' figure, not a measurement by AURION.
Indonesia
- Indonesian support
- not stated
- Regional languages
- not in the notes
- Jakarta hosting
- unknown
- Languages
- not in the notes
Where this comes from
Indonesia fields are carried on every model; unknown is an allowed value.
Industry Use Cases
No industry use case links this model.
Links
DeveloperXiaomibig tech, China
Xiaomibig tech, ChinaModelMiMo-V2-FlashXiaomi, LLM
MiMo-V2-FlashXiaomi, LLMR&D TopicsAgents and Tool UseEfficiency Architectures (MoE, Sparse Attention, Long Context)Open-Weight Frontier Releases as Research SubstrateReasoning via Large-Scale RL
Use CasesProcess Optimisation and Advanced Process ControlIndustrialClinical LLM (Open Weights)HealthcareTelecom Domain LLMs and BenchmarksTelecomChinese Medical LLMsHealthcareUse cases of models in the same R&D topic; none is recorded for this model itself.
Who Runs It
PT Amman Mineral Nusa Tenggara (AMMAN)In production, 8 Jul 2026MIND IDResearch, 14 Oct 2025
Conch CementIn production, 23 Jun 2025
PT Kalbe Morinaga IndonesiaIn production, Oct 2024+2 moreDeployments of these use cases; none names this model.
PT Amman Mineral Nusa Tenggara (AMMAN)In production, 8 Jul 2026MIND IDResearch, 14 Oct 2025Conch CementIn production, 23 Jun 2025PT Kalbe Morinaga IndonesiaIn production, Oct 2024+2 moreDeployments of these use cases; none names this model.Organisations 1
R&D topics 4
Open-Weight Frontier Releases as Research SubstrateUsing released weights as the base for downstream research.used byEfficiency Architectures (MoE, Sparse Attention, Long Context)Reducing training and inference cost per capability.produced byReasoning via Large-Scale RLReinforcement learning on outcomes, often verifiable math or code, to produce long reason…produced byAgents and Tool UseSystems that plan, call tools and act over multi-step tasks.produced by
Sources
- mimo.xiaomi.com/mimo-v2-flash https://mimo.xiaomi.com/mimo-v2-flash
- huggingface.co/XiaomiMiMo/MiMo-V2-Flash https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash
- artificialanalysis.ai/models/mimo-v2-0206 https://artificialanalysis.ai/models/mimo-v2-0206
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "MiMo-V2-Flash", https://aurion.semai.id/models/xiaomi-mimo-v2-flash/, records of October 2026.