Qt AI助手0.91发布——支持DeepSeek v3和3.7 Sonnet

Qt AI Assistant 0.91 Released – Support for DeepSeek v3 and 3.7 Sonnet

Qt AI助手0.91发布——支持DeepSeek v3和3.7 Sonnet

May 06, 2025 by Peter Schneider | Comments

​2025年5月6日：彼得·施耐德|评论

The latest release now supports two additional LLMs for code completion and prompts: DeepSeek v3 for our friends in China and 3.7 Sonnet for fans of Anthropic’s coding skills. The new release includes also an enhancement for the /fix functionality. 

最新版本现在支持两个额外的LLM用于代码完成和提示：DeepSeek v3适用于我们在中国的朋友，3.7 Sonnet适用于Anthropic编码技能的粉丝。新版本还包括对/fix功能的增强。

Applying of Fixed Code

固定代码的应用

It is now possible to apply changes suggested by LLMs directly to the code editor as a response to the /fix command.

现在可以将LLM建议的更改直接应用于代码编辑器，作为对/fix命令的响应。

DeepSeek v3 Support

DeepSeek v3支持

Based on popular requests from developers in China, we added DeepSeek v3 support. DeepSeek v3 can be used for code completion and prompts.

根据中国开发者的普遍要求，我们增加了对DeepSeek v3的支持。DeepSeek v3可用于代码补全和提示。

DeepSeek v3 scores an impressive 87% success rate for QML code completions, taking over the leadership position in the QML100FIM Benchmark. 

DeepSeek v3在QML代码完成方面取得了令人印象深刻的87%的成功率，在QML100FIM基准中占据了领先地位。

 Table: QML100FIM Code Completion Performance - May 2025

表：QML100FIM代码完成绩效-2025年5月

According to the QML100 Benchmark, it scores a 57% success rate for code generation through prompts. Do note that according to our tests, the underlying LLM DeepSeek v3 scores higher than the related reasoning model DeepSeek R1, which scores “only” 54% in prompt-based code generation.

根据QML100基准测试，通过提示生成代码的成功率为57%。请注意，根据我们的测试，底层LLM DeepSeek v3的得分高于相关的推理模型DeepSeek R1，后者在基于提示的代码生成中的得分“仅”为54%。

Claude 3.7 Sonnet Support

Claude 3.7Sonnet 支持

While QML coding performance stayed similar, we added support for 3.7 Sonnet for code completion and prompts.

虽然QML编码性能保持不变，但我们增加了对3.7 Sonnet的支持，用于代码完成和提示。

Claude 3.7 Sonnet maintains the leadership position with a 66% successful completion rate for prompt-based code completion.

Claude 3.7 Sonnet以66%的基于提示的代码完成率保持领先地位。

QML100 Prompt-based Coding Performance - May 2025

QML100基于提示的编码性能-2025年5月

Claude 3.7 Sonnet scores a 76% success rate for code completion while still sometimes struggling with correct indentation like its predecessor.

Claude 3.7 Sonnet的代码完成成功率为76%，但有时仍像其前身一样难以正确缩进。

How to Upgrade

如何升级

If you are already using the Qt AI Assistant, you can click the Update button in the Extensions view (remember that you need to upgrade first to Qt Creator 16.0.1).

如果已经在使用Qt AI助手，可以点击扩展视图中的更新按钮（记住，需要先升级到Qt Creator 16.0.1）。

If you are new to the Qt AI Assistant, you need to activate the use of external repositories in the extension view to fetch new extensions such as the Qt AI Assistant.

如果是Qt AI Assistant的新手，需要在扩展视图中激活外部存储库的使用，以获取新的扩展，如Qt AI Assistant。

Meanwhile… the following smaller enhancements have been made:

与此同时，已经进行了以下较小的增强：

• Linux on ARM support (experimental)
• ARM上的Linux支持（实验性）
• LLM error codes from OpenAI GPT displayed in General Messages
• 常规消息中显示的OpenAI GPT的LLM错误代码