Qwen3 Coder Next

Qwen3 Coder Next is an open-weight language model purpose-built for coding agents, developed by the Qwen Team at Alibaba. It uses a Mixture-of-Experts-style design with 80 billion total parameters but activates only 3 billion during inference, which gives it strong coding capability while keeping computational costs low. This efficiency-oriented architecture makes the model well suited for interactive software development tasks where responsiveness and reasoning quality both matter.

The model is trained through agentic methods that combine large-scale synthesis of verifiable coding tasks with executable environments, allowing it to learn directly from environment feedback during mid-training and reinforcement learning. Both base and instruction-tuned open-weight versions are released to support research and real-world coding agent development. On agent-centric benchmarks such as SWE-Bench and Terminal-Bench, Qwen3 Coder Next achieves competitive performance relative to its small active parameter count, making it a practical fit for developers who want capable coding assistance without the overhead of larger frontier models.