0.8B-9B native multimodal w/ more intelligence, less compute
Qwen3 is the large language model series developed by Qwen team, Alibaba Cloud. - QwenLM/Qwen3
Qwen3.5 Small supports 256K-token context windows extendable up to 1 million tokens, enabling users to handle ultra-long inputs like full codebases, research papers, or multi-turn conversations without truncation or context loss.
As an open-source model with publicly available weights, Qwen3.5 Small can be deployed locally using frameworks like Transformers, llama.cpp, Ollama, or vLLM. Users maintain full control over data, inference, and infrastructure without reliance on external APIs.
Qwen3.5 Small offers both thinking and non-thinking modes optimized for complex reasoning tasks. The thinking mode excels at mathematics, code generation, and logical reasoning, while non-thinking mode provides efficient general-purpose chat. Users can select the mode that best fits their use case and deploy locally or via inference frameworks.
Qwen3.5 Small supports post-training including supervised fine-tuning (SFT) and RLHF using frameworks like Axolotl and LLaMA-Factory, allowing teams to adapt the model to specialized vocabularies, domains, or behavioral requirements.
Qwen3.5 Small supports 100+ languages and dialects with strong multilingual instruction following and translation. It includes expert agent capabilities for precise tool integration in both thinking and non-thinking modes, enabling developers to build complex AI workflows.