Currently listed through these providers:
Model details
Qwen3.5-35B-A3B
Qwen3.5-35B-A3B is positioned as a native vision-language model within the Qwen 3.5 family, built around a hybrid architecture that combines Gated Delta Networks with a sparse Mixture-of-Experts design. This combination is intended to deliver high throughput and strong reasoning capability while keeping the active parameter count modest, making the model attractive for deployments where efficiency matters as much as raw quality. Its multimodal design allows it to accept both text and images as input while producing text outputs, supporting a range of reasoning and agentic workflows without requiring separate vision encoders.
The model serves as the foundation on which later Qwen releases build. In the Qwen 3.6 generation, the successor variant was highlighted as substantially outperforming Qwen3.5-35B-A3B on agentic coding tasks while remaining competitive with considerably larger dense models, signaling a meaningful step forward in coding capability and tool use. For practitioners, the original 3.5 release fits scenarios that need a balance of multimodal understanding, reasoning, and cost-effective inference, particularly when deployed through providers that expose the open weights for local or customized serving.
Quick Info
Powered by- Provider
- Weights & Biases
- Model key
- Qwen/Qwen3.5-35B-A3B
- Release date
- Feb 24, 2026
- Last updated
- Feb 24, 2026
- Input modalities
- Output modalities
- Capabilities
Cost
- Input token cost
- $0.25
- Output token cost
- $1.25
Limits
- Output tokens
- 262,144 tokens
- Context window
- 262,144 tokens