Qwen Code v0.12.5 Nightly Release

🔑 Enhanced Key Takeaways

• •Qwen Code is an AI-powered coding IDE integrating features like slash commands, MCP, PlanMode, TodoWrite, SubAgent support, multi-model switching, and tools such as WebFetch and Bash.[4]
• •Qwen3-Coder-Next, released by Alibaba's Qwen team in February 2026, serves as a key underlying model for Qwen Code, enabling advanced agentic coding with 256K context length.[1][2][3]
• •The Qwen Code roadmap includes upcoming features like Hooks, QwenCode SDK, extension system, LSP support, and integration with VSCode, ACP/Zed editors.[4]

🛠️ Technical Deep Dive

• •Qwen3-Coder-Next uses a hybrid architecture: Gated DeltaNet + MoE + Gated Attention, with 48 layers in a hybrid layout of 12 * (3 * (Gated DeltaNet -> MoE) -> 1 * (Gated Attention -> MoE)).[2]
• •MoE configuration: 512 total experts, 10 activated per token, 1 shared expert, expert intermediate dimension of 512.[2][3]
• •Technical specs: 80B total parameters (79B non-embedding), 3B activated, hidden dimension 2048, 262,144 native context length, trained via large-scale executable task synthesis + RL.[1][2][3]

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline