35B total / 3B active (MoE)
262,144 tokens
Apache-2.0
Official open-weight language world model release built on Qwen3.5-35B-A3B-Base and trained specifically for agentic environment simulation rather than general assistant behavior. Qwen describes it as the first language world model to cover seven agent interaction domains within a single model: MCP tool calling, Search, Terminal, SWE, Android, Web, and OS. The Hugging Face model card says the training pipeline runs from continual pretraining through supervised fine-tuning to RL, using world modeling as the objective from the CPT stage onward, and the accompanying paper reports more than 10 million environment-interaction trajectories across seven real-world domains plus a new AgentWorldBench evaluation benchmark. Release materials position the model both as a controllable environment simulator for agent training and as a foundation model whose world-model warmup improves downstream agent performance.