Key takeaways

• Qwen3 Coder is a Mixture-of-Experts (MoE) code generation model (480B parameters, 35B active) designed for agentic coding tasks like function calling, tool use, and long-context reasoning over repositories.
• The model demonstrates strong functional Python code generation (92.68% accuracy on a Python Code Generation Benchmark), effectively handling JSON output and tool utilization.
• A significant limitation is the model's struggle with numerical/mathematical precision and complex logical implementations, often adhering to inefficient solutions.
• The model exhibits poor performance on planning and multi-step execution tasks, with only 16.25% accuracy on Terminal-Bench due to dependency management issues, frequent timeouts, and environmental challenges.
• Qwen3 Coder scored only 4.56% on the Multi-Agent Counterfactual Reasoning Benchmark, highlighting weaknesses in complex, multi-step problem-solving and causal reasoning.

Technical specifications

Context length

1M tokens

Max output

—

Architecture

MoE

Modalities

text->text

Parameters

~175B

Open weights

Yes

Licence

Proprietary

Regions

China

Vendor

Qwen

Provider model key

qwen/qwen3-coder