Inference model · QWEN

Qwen 3.6 27B

The default. A capable dense model for coding, tool use, and structured agent loops.

Qwen 3.6 27B is the workhorse model in the catalog. Dense (no MoE routing surprises), 27B parameters, 128K context. Good at instruction following, function calling, and the kind of multi-step reasoning that agentic loops live on.

When to pick it

  • Default model for new agents that don’t yet know what they need
  • Code generation, code review, refactor agents
  • Tool-using agents where stable output schemas matter
  • Anything where you want a predictable cost-per-token curve

When to look elsewhere

  • Long-context summarization beyond 128K → look at Qwen 3.5 122B A10B (256K)
  • Very high throughput batch inference → DeepSeek V4 Flash
  • Cheap exploration of large-MoE behavior → Qwen 3.6 35B A3B