English

Improved

• Automatically upgrade to the bundled core on launch: If the installed core version is older than the version bundled with the app, CPA now automatically installs the bundled core and restores its runtime state (previously, this only ran when no core was installed).
• Renamed “Advanced Settings” to “Advanced Features” across navigation and API Access.

Fixed

• Fixed app updates potentially using cached responses or stale task state by adding cache-busting headers to update requests and refreshing release metadata before starting a download.

Included from v0.3.3

Added

• Added a separate proxy URL setting for each API connection in API Access and Easy Mode. It supports HTTP, HTTPS and SOCKS5 proxies.
• Model prices in Usage Records can now be fetched from Models.dev or LiteLLM. Preview matches, edit prices, and select which models to save before applying an update. Existing manually entered prices are unselected by default.

Improved

• Updated bundled model prices with entries for GPT-6 Sol, GPT-6 Luna, Grok 4.7, and more.
• Refined cost estimates for cache reads, cache creation, and long contexts. Missing cache prices are no longer inferred automatically.
• Updated the bundled core to 7.3.17.

简体中文

优化

• 启动时自动升级内置内核：应用启动时若检测到本地已安装的内核版本低于当前安装包内置版本，会自动升级至内置内核并恢复运行状态（此前仅在缺失内核时自动安装）。
• 界面文案调整：将导航栏及 API 接入中的“高级设置”统一更名为“高级功能”。

修复

• 修复软件更新可能受缓存影响或沿用过期任务状态的问题：为更新检查与清单请求增加防缓存机制，并在开始下载前重新刷新最新版本信息。

v0.3.3 更新内容

新增

• API 接入和简易模式新增每个 API 独立的代理 URL 设置，支持 HTTP、HTTPS、SOCKS5 代理。
• 使用记录的价格统计可从 Models.dev 或 LiteLLM 获取模型价格。更新前可预览匹配结果、编辑价格并选择要保存的模型；已有手动价格默认不勾选。

优化

• 更新内置模型价格，新增 GPT-6 Sol、GPT-6 Luna、Grok 4.7 等模型的价格数据。
• 调整成本估算中的缓存读取、缓存创建和长上下文计价；未提供的缓存价格不再自动推算。
• 内置内核更新至 7.3.17。