ApX logoApX logo

MiMo V2.6 Flash RL

Active Parameters

309B

Context Length

1.05M

Modality

Text

Architecture

Mixture of Experts

KV Cache Rate

144 KB/tok

License

MIT License

Release Date

21 Sept 2026

Knowledge Cutoff

-

API Pricing (per 1M)

Self-hosted only

Hardware & System Requirements

Precision

Native

Context

Workload Memory: 607.12 GB

FP16 · 8K context

Base Weights: 603.50 GB

KV Cache: 1.13 GB

Runtime & Graphs: 2.50 GB

Minimum Required Hardware

8x NVIDIA H200 141GB

607.1 / 1015.2 GB usable · TP 8

Determine requirements for your specific workload:

Upgrade to Pro

Runtime requirements

Checkpoint format

FP16

CUDA

>= 11.8

PyTorch

>= 2.1

Transformers

>= 5.3.0

ArchitecturePrecisionAttention kernel

Blackwell, Hopper

Native FP16

FlashAttention-3

Ada Lovelace, Ampere

Native FP16

FlashAttention-2

Architecture Diagram

Input TokensToken EmbeddingPosition: RoPEHidden: 4.1k · Context: 1.05M · Vocab: 152.6kx 48 layersRMSNormPre-AttentionSliding-Window Attention64Q / 4KV heads · SW: 128Head dim: 192+RMSNormPre-FFNSparse MoE FFN (8/256 experts)SwiGLUIntermediate: 2k+Final RMSNormOutput Logits

Evaluation Benchmarks

No evaluation benchmarks for MiMo V2.6 Flash RL available.

Rankings

Overall Rank

-

Coding Rank

-

About MiMo V2.6 Flash RL

MiMo V2.6 Flash RL is a high-efficiency reasoning model developed by Xiaomi, post-trained with reinforcement learning for accelerated reasoning and tool-use capabilities. It provides rapid response generation suitable for real-time mobile and edge assistant scenarios.

Specifications

Attention

Attention Structure

Single-Head Attention

Attention Heads

64

Key-Value Heads

4

Attention Head Dimension

192

Position Embedding

ROPE

RoPE Theta

10,000,000

Sliding Window Attention

Yes

Sliding Window Size

128

Sliding Window Ratio

81.3%

Normalization

RMS Normalization

Activation Function

SwigLU

Dimensions

Base Precision

FP16

Hidden Dimension Size

4,096

Number of Layers

48

FFN Intermediate Size (Dense)

16,384

Multi-Token Prediction Heads

3

Tokenizer

Vocabulary Size

152,576

Mixture of Experts

Total Expert Parameters

15.0B

Number of Experts

256

Active Experts

8

FFN Intermediate Size (per Expert)

2,048

Dense Layers Before MoE

1

About MiMo V2

MiMo-V2-Flash is a Mixture-of-Experts (MoE) model with hybrid attention architecture designed for high-speed reasoning and agentic workflows. It features Multi-Token Prediction (MTP) to achieve state-of-the-art performance while significantly reducing inference costs. The model is optimized for long-context modeling and efficient inference.


Other MiMo V2 Models