Qwen · 2025-09 · Sparse MoE
Qwen3-Next 80B A3B
Qwen3-Next 80B A3B is a grouped-query attention (GQA) transformer released by Qwen in 2025-09, with 48 layers, hidden size 2048 and a context window of 262,144 tokens.
Layer stack
Linear ×36 Attention ×12 · Layers 48
Key facts
| Family | Qwen |
|---|---|
| Released | 2025-09 |
| Params | 78.7 B |
| Active | 2.9 B |
| Context | 262,144 tokens |
| Attention | GQA (16:2) |
| Layers | 48 |
| Hidden | 2,048 |
| Heads | 16 |
| Vocab | 151,936 |
| Position | RoPE |
| Norm | RMSNorm |
| Activation | silu |
| Dtype | bf16 |
| Architecture class | Qwen3NextForCausalLM |
Architecture overview
Attention
grouped-query attention (GQA) — 16 q-heads / 2 kv-heads. RoPE θ=10,000,000.
Feed-forward / MoE
It is a sparse mixture-of-experts with 512 routed experts and top-10 routing; about 2.9B parameters are active per token.
Field-level comparison
Compared against the previous model of the same lab; where none exists, against the structurally closest model. The ratio column is this model divided by the comparison model.
Full-field comparison vs predecessor Qwen3 8B
| Model | Qwen3-Next 80B A3B | Qwen3 8B | Ratio |
|---|---|---|---|
| model_type | qwen3_next | qwen3 | ≠ |
| architectures | Qwen3NextForCausalLM | Qwen3ForCausalLM | ≠ |
| hidden_size | 2048 | 4096 | ×0.500 |
| num_hidden_layers | 48 | 36 | ×1.33 |
| num_attention_heads | 16 | 32 | ×0.500 |
| num_key_value_heads | 2 | 8 | ×0.250 |
| head_dim | 256 | 128 | ×2.00 |
| intermediate_size | 5120 | 12288 | ×0.417 |
| hidden_act | silu | silu | ＝ |
| num_experts_per_tok | 10 | — | |
| moe_intermediate_size | 512 | — | |
| rope_theta | 10000000 | 1000000 | ×10.00 |
| max_position_embeddings | 262144 | 40960 | ×6.40 |
| rms_norm_eps | 0.000001 | 0.000001 | ≈1 |
| vocab_size | 151936 | 151936 | ≈1 |
| torch_dtype | bfloat16 | bfloat16 | ＝ |
| attention_dropout | 0 | 0 | ＝ |
| bos_token_id | 151643 | 151643 | ≈1 |
| decoder_sparse_step | 1 | — | |
| eos_token_id | 151645 | 151645 | ≈1 |
| full_attention_interval | 4 | — | |
| initializer_range | 0.02 | 0.02 | ≈1 |
| linear_conv_kernel_dim | 4 | — | |
| linear_key_head_dim | 128 | — | |
| linear_num_key_heads | 16 | — | |
| linear_num_value_heads | 32 | — |
Most similar architectures
- Qwen3-Coder-Next (Qwen, 2026-01) 1.000
- Qwen3.8-Flash-Next (Qwen, 2026-08) 1.000
- Qwen3.6 35B A3B (Qwen, 2026-04) 0.999
- Qwen3 30B A3B (Qwen, 2025-04) 0.997
- Hunyuan-TurboS 30B A3B (Hunyuan, 2026-05) 0.997
- GPT-OSS 120B (OpenAI, 2025-08) 0.996
Similarity values range from 0 (no shared categorical features) to 1 (identical profiles).
Raw config fields
37 fields
| architectures | Qwen3NextForCausalLM |
|---|---|
| attention_dropout | 0 |
| bos_token_id | 151643 |
| decoder_sparse_step | 1 |
| eos_token_id | 151645 |
| full_attention_interval | 4 |
| head_dim | 256 |
| hidden_act | silu |
| hidden_size | 2048 |
| initializer_range | 0.02 |
| intermediate_size | 5120 |
| linear_conv_kernel_dim | 4 |
| linear_key_head_dim | 128 |
| linear_num_key_heads | 16 |
| linear_num_value_heads | 32 |
| linear_value_head_dim | 128 |
| max_position_embeddings | 262144 |
| model_type | qwen3_next |
| moe_intermediate_size | 512 |
| norm_topk_prob | true |
| num_attention_heads | 16 |
| num_experts | 512 |
| num_experts_per_tok | 10 |
| num_hidden_layers | 48 |
| num_key_value_heads | 2 |
| output_router_logits | false |
| partial_rotary_factor | 0.25 |
| rms_norm_eps | 0.000001 |
| rope_theta | 10000000 |
| router_aux_loss_coef | 0.001 |
| shared_expert_intermediate_size | 512 |
| tie_word_embeddings | false |
| torch_dtype | bfloat16 |
| transformers_version | 4.57.0.dev0 |
| use_cache | true |
| use_sliding_window | false |
| vocab_size | 151936 |