Skip to main content

July 2025 product updates

Qoder now supports the Qwen3-Coder model, delivering significant improvements in code accuracy and response speed. You can also configure multiple AI rules in a project.

Qwen3-Coder model support

This new model achieves state-of-the-art (SOTA) performance for code generation among open-source models. It features powerful coding agent capabilities, excels at tool calling and environment interaction, and supports a context length of up to 1 million tokens. 
} QWEN3-CODER
                                        Open Models                    Proprietary Models
 Benchmarks              Qwen3-Coder    Kimi-K2   DeepSeek-V3    Claude     OpenAI
                       480B-A35B-Instruct Instruct    0324      Sonnet-4    GPT-4.1
                              Agentic Coding
 Terminal-Bench              37.5        30.0        2.5        35.5       25.3
 SWE-bench Verified          69.6          -          -         70.4         -
   w/ OpenHands, 300 turns   67.0        65.4       38.0        68.0       45.6
   w/ OpenHands, 100 turns     -         65.3          -        72.7       63.8
   w/ Private Scaffolding      -           -           -          -          -
 SWE-bench Lite              24.3        23.3       13.0        27.7         -
 SWE-bench Multilingual      54.7        47.3       13.0        51.3       31.5
 Multi-SWE-bench mini        29.0        13.8        7.5          -          -
 Multi-SWE-bench Flash       25.0        29.7          -        25.0         -
 Aider-Polyglot              61.8        65.2       54.2        51.4         -
 Spider2                     31.1        22.2       12.8        31.1       14.5
                           Agentic Browser Use
 WebArena                   69.9        47.4       40.0        51.1       44.3
 Mind2Web                   55.8        42.7       36.0        47.4       49.6
                            Agentic Tool Use
 BFCL-v3                    68.7        63.2       56.9        73.3       62.9
 TAU-Bench Retail            77.5        35.2       59.1        88.5         -
 TAU-Bench Airline           60.0        53.5       40.0        60.0         -
The model selection in Tongyi Lingma now includes an Auto mode and the new Qwen3-Coder model, while the DeepSeek model option has been removed. We will continue to add the latest models to provide the best performance.

Enhancements

  • You can now configure multiple AI rules in a project.
  • Simplified header configuration for the MCP service.
  • You can now view modified files from a chat response.
  • Improved code completion performance.
  • Clarified error messages for path-related issues.
  • Optimized inline suggestion prediction and generation.
  • Clarified error messages for failed file modifications.
  • Clarified error messages for path-related issues.
  • Improved the fallback mechanism for inline code suggestions.