Model overview · Qwen3-Next
Qwen3-Next 80B-A3B
Generated from Architecture IR 0.1.0 · revision b664bd1e89
· report a correction
Overview
Fact sheet
- Layers
- 48
- Hidden size
- 2048
- Attention heads
- 16 · KV 2
- Vocab
- 151,936
- Context
- 256K
- Params
- 80B total / 3B active
Architecture Genome
One column per layer — top: attention, bottom: FFN/MoE.
Layer
attention (GQA/MLA) linear / recurrent attention MoE dense FFN
Evidence
Every visible number resolves to one of these claims. Inferred or unknown values are never shown as fact.
| Claim | Value | Status | Source |
|---|---|---|---|
facts.num_hidden_layers | 48 | • reported | hf_config · $.text_config.num_hidden_layers |
facts.hidden_size | 2048 | • reported | hf_config · $.text_config.hidden_size |
facts.num_attention_heads | 16 | • reported | hf_config · $.text_config.num_attention_heads |
facts.num_key_value_heads | 2 | • reported | hf_config · $.text_config.num_key_value_heads |
facts.head_dim | 256 | • reported | hf_config · $.text_config.head_dim |
facts.vocab_size | 151936 | • reported | hf_config · $.text_config.vocab_size |
facts.context_tokens | 262144 | • reported | hf_config · $.text_config.max_position_embeddings |
facts.total_params | 80000000000 | • reported | model_card · parameter count |
facts.active_params | 3000000000 | = derived | derived · activated experts + attention params |
topology.attention.full_attention_interval | 4 | • reported | hf_config · $.text_config.full_attention_interval |
topology.attention_groups[0] | "36 Gated DeltaNet layers (3 of every 4)" | • reported | hf_config · $.text_config.full_attention_interval |
topology.attention_groups[1] | "12 Gated Attention layers (1 of every 4)" | • reported | hf_config · $.text_config.full_attention_interval |
topology.experts.routed_total | 512 | • reported | hf_config · $.text_config.num_experts |
topology.experts.active_routed | 10 | • reported | hf_config · $.text_config.num_experts_per_tok |
Compare
Compare this model with any other catalog model (see catalog).