Qwen · 2026-02 · Sparse MoE · Multimodal
Qwen3.5 122B A10B
Qwen3.5 122B A10B is a grouped-query attention (GQA) transformer released by Qwen in 2026-02, with 48 layers, hidden size 3072 and a context window of 262,144 tokens.
Layer stack
Linear ×36 Attention ×12 · Layers 48
Key facts
| Family | Qwen |
|---|---|
| Released | 2026-02 |
| Params | 119.9 B |
| Active | 7.6 B |
| Context | 262,144 tokens |
| Attention | GQA (32:2) |
| Layers | 48 |
| Hidden | 3,072 |
| Heads | 32 |
| Vocab | 248,320 |
| Position | learned/absolute |
| Norm | RMSNorm |
| Activation | silu |
| Dtype | bf16 |
| Architecture class | Qwen3_5MoeForConditionalGeneration |
Architecture overview
Attention
grouped-query attention (GQA) — 32 q-heads / 2 kv-heads.
Feed-forward / MoE
It is a sparse mixture-of-experts with 256 routed experts and top-8 routing; about 7.6B parameters are active per token.
The configuration ships 1 extra multi-token-prediction (MTP) layers used to accelerate decoding.
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-Coder-Next
| Model | Qwen3.5 122B A10B | Qwen3-Coder-Next | Ratio |
|---|---|---|---|
| model_type | qwen3_5_moe | qwen3_next | ≠ |
| architectures | Qwen3_5MoeForConditionalGeneration | Qwen3NextForCausalLM | ≠ |
| hidden_size | 3072 | 2048 | ×1.50 |
| num_hidden_layers | 48 | 48 | ≈1 |
| num_attention_heads | 32 | 16 | ×2.00 |
| num_key_value_heads | 2 | 2 | ≈1 |
| head_dim | 256 | 256 | ≈1 |
| intermediate_size | — | 5120 | |
| hidden_act | silu | silu | ＝ |
| num_experts_per_tok | 8 | 10 | ×0.800 |
| moe_intermediate_size | 1024 | 512 | ×2.00 |
| rope_theta | — | 5000000 | |
| max_position_embeddings | 262144 | 262144 | ≈1 |
| rms_norm_eps | 0.000001 | 0.000001 | ≈1 |
| vocab_size | 248320 | 151936 | ×1.63 |
| torch_dtype | — | bfloat16 | |
| image_token_id | 248056 | — | |
| tie_word_embeddings | false | false | ＝ |
| transformers_version | 4.57.0.dev0 | 4.57.0.dev0 | ＝ |
| video_token_id | 248057 | — | |
| vision_end_token_id | 248054 | — | |
| vision_start_token_id | 248053 | — | |
| attention_bias | false | false | ＝ |
| attention_dropout | 0 | 0 | ＝ |
| attn_output_gate | true | — | |
| dtype | bfloat16 | — |
Most similar architectures
- Qwen3.5 35B A3B (Qwen, 2026-02) 0.998
- Qwen3.5 397B A17B (Qwen, 2026-02) 0.997
- Hunyuan 3 Preview (Hunyuan, 2026-04) 0.993
- Hunyuan 3 Preview Base (Hunyuan, 2026-04) 0.993
- Hunyuan 3 (Hunyuan, 2026-07) 0.993
- Qwen3.8-Flash-Next (Qwen, 2026-08) 0.926
Similarity values range from 0 (no shared categorical features) to 1 (identical profiles).
Raw config fields
39 fields
| architectures | Qwen3_5MoeForConditionalGeneration |
|---|---|
| image_token_id | 248056 |
| model_type | qwen3_5_moe |
| tie_word_embeddings | false |
| transformers_version | 4.57.0.dev0 |
| video_token_id | 248057 |
| vision_end_token_id | 248054 |
| vision_start_token_id | 248053 |
| attention_bias | false |
| attention_dropout | 0 |
| attn_output_gate | true |
| dtype | bfloat16 |
| eos_token_id | 248044 |
| full_attention_interval | 4 |
| head_dim | 256 |
| hidden_act | silu |
| hidden_size | 3072 |
| initializer_range | 0.02 |
| layer_types | linear_attention×36 + full_attention×12 |
| linear_conv_kernel_dim | 4 |
| linear_key_head_dim | 128 |
| linear_num_key_heads | 16 |
| linear_num_value_heads | 64 |
| linear_value_head_dim | 128 |
| max_position_embeddings | 262144 |
| moe_intermediate_size | 1024 |
| mtp_num_hidden_layers | 1 |
| mtp_use_dedicated_embeddings | false |
| num_attention_heads | 32 |
| num_experts | 256 |
| num_experts_per_tok | 8 |
| num_hidden_layers | 48 |
| num_key_value_heads | 2 |
| rms_norm_eps | 0.000001 |
| router_aux_loss_coef | 0.001 |
| shared_expert_intermediate_size | 1024 |
| use_cache | true |
| vocab_size | 248320 |
| mamba_ssm_dtype | float32 |