Model techniques map
Techniquesmodel architecturechannel mixermixture of experts

general family · filed under model architecture

Mixture of Experts

A model architecture using a mixture of experts; the evidence does not specify a more particular routing or activation mechanism.

Also called 198B total params / 11B activated params, 552B-parameter MoE, Mixture of Experts (MoE), Mixture-of-Experts (MoE) architecture, Mixture-of-Experts (MoE) architectures, MoE.

sources
88
models
24
labs adopt it
13
strongest
core

How sources treat it

One count per evidence span, weakest treatment to strongest.

used 6core 95

Documented in

Further reading

Picked by hand, not extracted: where to read more, not evidence for anything on this page.

In the code

Checked as “Mixture of experts” against each model’s own modeling code and config.json, independently of its documents. A model’s page shows the quoted lines.

ModelCodeIts documentsWhat the code does
GLM-5.3-Flash in its code coreGlm5NextTextMoE routes each token to num_experts_per_tok=8 of n_routed_experts=288 expert MLPs via a learned sigmoid router (Glm5NextTextTopkRouter) in sparse mlp_layer_types layers. modeling_glm5_next.py#L178
DeepSeek-V4.1-Flash in its code coreFeed-forward layers (DeepseekV4MoE) route each token to a learned top-k subset of n_routed_experts routed experts. model.py#L118
Hy4-preview in its code coreHYV4TopkRouter routes each token to num_experts_per_tok of n_routed_experts expert MLPs via sigmoid-scored top-k selection, used in every layer where mlp_layer_types is "sparse". modeling_hy_v4.py#L543
DeepSeek-V4-Flash-0731 in its codenot statedFeed-forward layers use DeepseekV4SparseMoeBlock with a learned TopKRouter (or hash router) selecting num_experts_per_tok=6 of n_routed_experts=256 experts per token. modeling_deepseek_v4.py#L1086
NVIDIA-Nemotron-3-Ultra-550B-A55B in its code coreMoE layers route tokens to a subset of 512 routed experts via a top-k router selecting 22 experts per token. modeling_nemotron_h.py#L730
MiMo-V2.6-Flash in its code coreMoE feed-forward layers route each token via a sigmoid-scored top-8-of-256 router (MiMoV2MoEGate/MiMoV2MoE) applied on layers flagged in moe_layer_freq. modeling_mimo_v2.py#L191
DeepSeek-V4-Flash in its code coreFeed-forward layers route tokens to top num_experts_per_tok=6 of n_routed_experts=256 experts via a learned TopKRouter (or hash router for early layers). modeling_deepseek_v4.py#L1086
MiMo-V2.5 in its code coreMoE layers route each token via a sigmoid-scored router (noaux_tc) to 8 of 256 routed experts per token, per moe_layer_freq. modeling_mimo_v2.py#L191
GLM-5.3 in its codenot statedSparse MLP layers route each token via a learned TopkRouter to 8 of 256 routed experts (n_routed_experts=256, num_experts_per_tok=8). modeling_glm_moe_dsa.py#L513
Hy3 in its code coreFeed-forward layers (all but the first) route tokens to 8 of 192 experts via a learned sigmoid router modeling_hy_v3.py#L384
GLM-5.2 in its code coreMoE decoder layers route each token via GlmMoeDsaTopkRouter to 8 of 256 routed experts (GlmMoeDsaExperts) on layers marked 'sparse'. modeling_glm_moe_dsa.py#L592
MiniMax-M3 in its code coreFeed-forward layers marked 'sparse' in mlp_layer_types route each token to num_experts_per_tok of num_local_experts via a sigmoid top-k router. modeling_minimax_m3_vl.py#L243
DeepSeek-V3.2 in its codenot statedDeepseekV32MoE routes tokens to n_routed_experts=256 experts via a top-k sigmoid router (num_experts_per_tok=8) in every layer beyond first_k_dense_replace. modeling_deepseek_v32.py#L542
DeepSeek-V4-Flash-Vision-Exp in its code coreFeed-forward layers route tokens to a subset of n_routed_experts=256 expert MLPs with num_experts_per_tok=6 chosen by a learned TopKRouter (or hash router for early layers). modeling_deepseek_v4.py#L1086
DeepSeek-V4-Pro in its code coreFeed-forward MoE layers route tokens to top-6 of 384 routed experts via a learned router modeling_deepseek_v4.py#L1086
DeepSeek-V4-Pro-0813 in its codenot statedThe SparseMoeBlock routes tokens to a top-k of n_routed_experts=384 expert MLPs via a learned router (TopKRouter or HashRouter for early layers). modeling_deepseek_v4.py#L1086
Gemma 4 31B not in its code coreMoE code (router + experts) exists but enable_moe_block is false and num_experts/top_k_experts are null in this config, so no routing occurs. modeling_gemma4.py#L1376
Inkling in its code coreFeed-forward layers past dense_mlp_idx route each token to a top-k subset of 256 routed experts chosen by a sigmoid gate. moe.py#L442
Kimi K3 in its code coreKimiSparseMoeBlock routes tokens among num_experts=896 expert MLPs choosing num_experts_per_token=16 via a learned KimiMoEGate router, applied to layers past first_k_dense_replace. modeling_kimi_linear.py#L786
Laguna-S-2.1 in its code coreLagunaSparseMoeBlock routes tokens to top-10 of 256 experts via a learned sigmoid router, applied in layers marked sparse. modeling_laguna.py#L242
MiMo-V2.5-Pro in its code coreFeed-forward layers route tokens to 8-of-384 experts chosen by a learned sigmoid router (noaux_tc top-k selection). modeling_mimo_v2.py#L186
MiMo-V2.6-Pro in its code coreMoE feed-forward layers route tokens to 8 of 384 experts per token via a learned sigmoid router (MiMoV2MoEGate), active per moe_layer_freq. modeling_mimo_v2.py#L191
NVIDIA-Nemotron-3.5-Lightning-30B-A3B in its code coreNemotronHTopkRouter routes each token to num_experts_per_tok=6 of n_routed_experts=128 expert MLPs via sigmoid scores and top-k selection. modeling_nemotron_h.py#L759
Qwen3.5-397B-A17B in its code coreEach MoE block routes tokens via a learned TopKRouter to a subset (10 of 512) of expert MLPs. modeling_qwen3_5_moe.py#L897
Qwen3.6-35B-A3B in its code coreQwen3_5MoeSparseMoeBlock routes each token to num_experts_per_tok=8 of num_experts=256 expert MLPs via a learned TopKRouter. modeling_qwen3_5_moe.py#L897
Qwen3.8-Flash-Next in its code coreQwen4ExpTextSparseMoeBlock routes each token via a top-k router (Qwen4ExpTextTopKRouter) to a subset of the 512 routed experts. modeling_qwen4_exp.py#L973
Step-3.7-Flash in its code coreStep3p7SparseMoeBlock routes tokens via Step3p7TopKRouter's top-k selection over 288 routed experts with 8 selected per token. modeling_step3p7.py#L606
gpt-oss-120b in its code coreEach token is routed via top-k (4 of 128) learned router to expert MLPs. modeling_gpt_oss.py#L128

Evidence

101 spans quoted from the sources, strongest treatment first.

MiMo-V2.6-Pro is a Mixture of Experts (MoE) model with 1.0 trillion total parameters, but only 42 billion active parameters are used during inference.

coremodel architecturein MiMo-V2.6-ProXiaomi

Inkling is a mixture-of-experts system with 975 billion total parameters

coremodel architecturein InklingThinking Machines Lab

NVIDIA Nemotron 3.5 Lightning is an open 30B mixture-of-experts (MoE) model with 3B active parameters built for that execution layer of always-on agents.

coremodel architecturein Nemotron 3.5 LightningNVIDIA

Architecture: Mixture-of-Experts (MoE)

coremodel architecturein Kimi K3Moonshot AI

Qwen3.5-397B-A17B is a 397 billion parameter Mixture-of-Experts model with only 17 billion active parameters per token.

coremodel architecturein Qwen3.5-397B-A17BAlibaba

Hy3 is a 295B-parameter Mixture-of-Experts (MoE) model

coremodel architecturein Hy3Tencent Hunyuan

a 198B-parameter sparse Mixture-of-Experts (MoE) vision-language model

coremodel architecturein Step 3.7 FlashStepFun

a 198B-parameter sparse Mixture-of-Experts (MoE) vision-language model

coremodel architecturein Step 3.7 FlashStepFun

scaled up Mixture of Experts (MoE) sparsity, effectively activating 16 out of 896 experts

coremodel architecturein Kimi K3Moonshot AI

Architecture: Sparse MoE (Mixture of Experts), 1.02T total / 42B activated parameters

coremodel architecturein MiMo-V2.6-Pro-RLXiaomi

native omnimodal MoE model

coremodel architecturein MiMo-V2.5Xiaomi

DeepSeek V4 is a Mixture-of-Experts (MoE) language model series

coremodel architecturein DeepSeek-V4DeepSeek

breakthrough hybrid mixture-of-experts architecture

coremodel architecturein Nemotron 3 familyNVIDIA

198B total params / 11B activated params

coremodel architecturein Step 3.7 FlashStepFun

The gpt-oss models are autoregressive Mixture-of-Experts (MoE) transformers

coremodel architecturein gpt-oss-120b and gpt-oss-20bOpenAI

open-weight multimodal mixture-of-experts model

coremodel architecturein InklingThinking Machines Lab

The backbone consists of 78 layers, where the first layer uses a standard dense FFN and the remaining 77 layers replace it with MoE

coremodel architecturein Hy4-previewTencent Hunyuan

The backbone consists of 78 layers, where the first layer uses a standard dense FFN and the remaining 77 layers replace it with MoE, each containing 256 routed experts and 1 shared expert

coremodel architecturein Hy4-previewTencent Hunyuan

Featuring both Dense and Mixture-of-Experts (MoE) architectures

coremodel architecturein Gemma 4Google DeepMind

Hy3 is a 295B-parameter Mixture-of-Experts (MoE) model.

coremodel architecturein Hy3Tencent

Hy3’s architecture contains a sparse MoE with 192 experts and top-8 routing.

coremodel architecturein Hy3Tencent

109B-parameter Mixture of Experts (MoE) model trained from scratch

coremodel architecturein MiniMax-M3MiniMax

with approximately 550 billion total parameters and 55 billion active

coremodel architecturein Nemotron 3 UltraNVIDIA

A 66-layer decoder-only transformer with a sparse Mixture-of-Experts (MoE) feed-forward backbone

coremodel architecturein InklingThinking Machines Lab

Hy3 is a 295B-parameter Mixture-of-Experts (MoE) model

coremodel architecturein Hy3Tencent Hunyuan

activates just 16 of its 896 experts per token

coremodel architecturein Kimi K3Moonshot AI

550 billion total and 55 billion active parameter Mixture-of-Experts

coremodel architecturein Nemotron 3 UltraNVIDIA

Nemotron 3 Ultra : Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning

coremodel architecturein Nemotron 3 UltraNVIDIA

Nemotron 3 Ultra uses a MoE Hybrid Mamba-Attention architecture

coremodel architecturein Nemotron 3 UltraNVIDIA

Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning

coremodel architecturein Nemotron 3 UltraNVIDIA

It is a mixture-of-experts model with 320B total parameters and 18B active per token

coremodel architecturein GLM-5.3-FlashZ.ai

Inkling is "a Mixture-of-Experts transformer with 975B total parameters, 41B active"

coremodel architecturein InklingThinking Machines Lab

MoE experts: 512 experts, 10 routed + 1 shared activated per token

coremodel architecturein Qwen3.8-Flash-NextQwen

The model employs a hybrid Latent Mixture-of-Experts (LatentMoE) architecture, utilizing interleaved Mamba-2 and MoE layers

coremodel architecturein Nemotron 3 UltraNVIDIA

It is a Mixture-of-Experts (MoE) model with 8B activated parameters per token.

coremodel architecturein Laguna S 2.1Poolside

MiMo-V2.5-Pro is an open-source Mixture-of-Experts (MoE) language model

coremodel architecturein MiMo-V2.5-ProXiaomi

Qwen3.8-Flash-Next is a sparse mixture-of-experts model

coreunclearin Qwen3.8-Flash-NextQwen

MiMo-V2.5-Pro is a Mixture of Experts (MoE) model with 1.0 trillion total parameters, but only 42 billion active parameters are used during inference.

coremodel architecturein MiMo-V2.5-ProXiaomi

Employs Mixture-of-Experts Hybrid Mamba-Attention architecture.

coremodel architecturein Nemotron 3 UltraNVIDIA

552B-parameter MoE.

coremodel architecturein DeepSeek-V4.1-FlashDeepSeek

NVIDIA Nemotron 3.5 Lightning is an open mixture-of-experts model from NVIDIA

coreunclearin Nemotron 3.5 LightningNVIDIA

Inkling is a Mixture-of-Experts transformer with 975B total parameters, 41B active.

coremodel architecturein InklingThinking Machines Lab

Hy4 preview is a mixture-of-experts model from Tencent

coreunclearin Hy4-previewTencent

Both models are mixture-of-experts: gpt-oss-120b activates 5.1B parameters per token

coremodel architecturein gpt-oss-120b and gpt-oss-20bOpenAI

Both big models are sparse mixture-of-experts

coreunclearin MiMo-V2.6-Pro and MiMo-V2.6-FlashXiaomi

we introduce DeepSeek-V4.1-Flash, a multimodal Mixture-of-Experts (MoE) model with support for contexts of up to one million tokens.

coremodel architecturein DeepSeek-V4.1-FlashDeepSeek

DeepSeek-V4-Pro carries 1.6 trillion parameters with 49 billion activated per token.

coremodel architecturein DeepSeek-V4DeepSeek

The MoE framework that manages 896 experts with 16 active per token.

coremodel architecturein Kimi K3Moonshot AI

Qwen3.8-Flash-Next is a multimodal, ultra-sparse Mixture-of-Experts model.

coremodel architecturein Qwen3.8-Flash-NextQwen

The reference inference covers the vision encoder and aligner, DFlash attention, MoE, Hyper-Connections, and the DSpark forward path.

coremodel architecturein DeepSeek-V4-Flash-Vision-ExpDeepSeek

integrates a linear attention mechanism with a sparse mixture-of-experts model

coremodel architecturein Qwen3.5-397B-A17BQwen

117B parameters with 5.1B active parameters

coremodel architecturein gpt-ossOpenAI

sparse mixture-of-experts

coremodel architecturein Qwen3.5-397B-A17BAlibaba

its Mixture-of-Experts (MoE) architecture means only approximately 40 billion parameters are active

coremodel architecturein GLM-5.2Z.ai

Inkling is a 975B-parameter sparse Mixture-of-Experts model with 41B active parameters

coremodel architecturein InklingThinking Machines Lab

DeepSeek-V4-Pro: 1.6T total / 49B active params

coremodel architecturein DeepSeek-V4DeepSeek

a sparse Mixture-of-Experts (MoE) feed-forward backbone

coremodel architecturein InklingThinking Machines Lab

117B-parameter Mixture-of-Experts (MoE) language model ... activates 5.1B parameters per forward pass

coremodel architecturein gpt-oss-120bOpenAI

multimodal mixture-of-experts model with approximately 321B total parameters and 18B active parameters per token

coremodel architecturein GLM-5.3-FlashZ.ai

Built on a Mixture of Experts (MoE) architecture

coremodel architecturein Hy3Tencent

Hy3 is a hybrid fast-and-slow-thinking model built on a Mixture-of-Experts (MoE) architecture

coremodel architecturein Hy3Tencent

Mixture-of-Experts language model

coremodel architecturein Hy4-previewTencent Hunyuan

two strong Mixture-of-Experts (MoE) language models

coremodel architecturein DeepSeek-V4-FlashDeepSeek

M.1 and XS.2 are Mixture-of-Experts foundation models.

coremodel architecturein LAGUNA M.1 and LAGUNA XS.2Poolside

Model type: Mixture of Experts (MoE)

coremodel architecturein Laguna S 2.1Poolside

In an MoE model: On the other hand, the router chooses only a small subset of experts.

coreunclearin Nemotron 3.5 LightningNVIDIA

Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model

coremodel architecturein Nemotron 3 UltraNVIDIA

GLM-5 scales to 256 experts and reduces its layer count to 80 to minimize expert parallelism communication overhead.

coremodel architecturein GLM-5Z.ai

while the MoE experts are compressed to W4A8 (INT4)

coremodel architecturein GLM-5Z.ai

It is a MoE model with a total of 295 billion parameters and 21 billion active parameters

coremodel architecturein Hy3Tencent

Mixture Of Experts - Number of Experts: 256 - Number of Activated Experts: 8 Routed + 1 Shared

coremodel architecturein Qwen3.6-35B-A3BQwen

Architecture: Sparse MoE (Mixture of Experts), 310B total / 15B activated parameters

coremodel architecturein MiMo-V2.5Xiaomi

Mixture-of-Experts Hybrid Mamba-Attention architecture

coremodel architecturein Nemotron 3 UltraNVIDIA

Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning

coremodel architecturein Nemotron 3 UltraNVIDIA

Nemotron 3 Ultra uses a MoE Hybrid Mamba-Attention architecture

coremodel architecturein Nemotron 3 UltraNVIDIA

two strong Mixture-of-Experts (MoE) language models — DeepSeek-V4-Pro with 1.6T parameters (49B activated) and DeepSeek-V4-Flash with 284B parameters (13B activated)

coremodel architecturein DeepSeek-V4DeepSeek

Built on a Mixture-of-Experts architecture with 309B total parameters and 15B activated per token

coremodel architecturein MiMo-V2.6-FlashXiaomi

Featuring both Dense and Mixture-of-Experts (MoE) architectures, Gemma 4 is well-suited for tasks like text generation, coding, and reasoning.

coremodel architecturein Gemma 4Google DeepMind

By only activating a 4B subset of parameters during inference, the Mixture-of-Experts model runs much faster than its 26B total might suggest.

coremodel architecturein Gemma 4Google DeepMind

Laguna S 2.1 is a 118B total parameter Mixture-of-Experts model with 8B activated parameters per token

coremodel architecturein Laguna S 2.1Poolside

Architecture: Sparse MoE (Mixture of Experts)

coremodel architecturein MiMo-V2.5Xiaomi

Architecture: Sparse MoE (Mixture of Experts), 310B total / 15B activated parameters

coremodel architecturein MiMo-V2.5Xiaomi

DeepSeek V4.1 Flash is a sparse mixture-of-experts model from DeepSeek

coremodel architecturein DeepSeek-V4.1-FlashDeepSeek

Step 3.7 Flash is a 198B-parameter sparse Mixture-of-Experts (MoE) vision-language model

coremodel architecturein Step 3.7 FlashStepFun

Step 3.7 Flash is a 198B-parameter sparse Mixture-of-Experts (MoE) vision-language model

coremodel architecturein Step 3.7 FlashStepFun

Architecture: Sparse MoE (Mixture of Experts), 309B total / 15B activated parameters

coremodel architecturein MiMo-V2.6-Flash-RLXiaomi

employs a sparse Mixture of Experts (MoE) architecture

coremodel architecturein Step 3.7 FlashStepFun

It is a sparse mixture-of-experts model with 13B active parameters out of 284B total.

coremodel architecturein DeepSeek-V4-Flash-Vision-ExpDeepSeek

Step 3.7 Flash is a 198B-parameter sparse Mixture-of-Experts (MoE) vision-language model

coremodel architecturein Step 3.7 FlashStepFun

MiMo-V2-Flash is a Mixture-of-Experts (MoE) model with 309B total parameters

coremodel architecturein MiMo-V2-FlashXiaomi

MoE models suffer from inconsistent expert routing across rollout and training due to numerical precision issues

coremodel architecturein MiMo-V2-FlashXiaomi

Mixture-of-Experts language model

coremodel architecturein Hy3Tencent

a multimodal MoE model

coremodel architecturein Qwen3.8-Flash-NextQwen

Laguna S 2.1 is a 118B total parameter Mixture-of-Experts (MoE) model with 8B activated parameters per token

coremodel architecturein Laguna S 2.1Poolside

MiMo-V2.5 is a 310B-parameter Sparse MoE model (15B active)

coremodel architecturein MiMo-V2.5Xiaomi

activating only 3.8 billion of its total parameters during inference to deliver exceptionally fast tokens-per-second

usedmodel architecturein Gemma 4 26BGoogle

The MoE contains 64 experts with top-4 expert routing

usedmodel architecturein MiniMax Sparse AttentionMiniMax

the Gemma 4 model suite features dense and Mixture-of-Experts architectures

usedmodel architecturein Gemma 4Google DeepMind

The MoE backbone likely stays.

usedmodel architecturein MiniMax-M3MiniMax

Nemotron 3 Ultra: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model

usedmodel architecturein Nemotron 3 UltraNVIDIA

The model runs on a Mixture-of-Experts (MoE) architecture with 1 trillion total parameters and 42 billion active parameters per inference pass.

usedmodel architecturein MiMo-V2.5-ProXiaomi

Filed alongside

Other methods under model architecture :: channel mixer :: mixture of experts.