jlcodes99/cockpit-tools v1.3.63 on GitHub

更新日志（中文）

新增 GPT-6.1 Sol 官方价格估算，覆盖缓存读取/写入、Fast/Flex 档位和长上下文费率。
内置模型目录、唤醒/鹈鹕测智选择、供应商测试候选和网关原生模型识别新增 GPT-6.1 Sol，保留已有默认模型选择；GPT-6.1 Sol 在唤醒任务和鹈鹕测智中支持 ultra 推理档位。
模型供应商提供“鹈鹕测智”入口，复用现有弹框，支持选择已保存的供应商、API Key 和模型；结果标明来源，共用历史和重试。测试使用独立临时网关，重试时重新读取已保存的凭据。

模型供应商对话测试的本地 Responses 网关与唤醒统一使用默认模型 gpt-5.6-luna，供应商目录包含该模型时优先选择；保留手动指定模型及供应商自身的上游模型。
手动刷新配额失败时现在会提示本地代理错误，不再没有任何反馈。
Windows 上扫描 Antigravity 元数据时不再探测 WSL 路径，避免顺带启动 WSL 或长时间卡住。
其他批次更新时鹈鹕测智结果不再自动跳走，重试失败项会继续停留在当前批次。
窄窗口下 Codex 账号工具栏按钮不再挤压搜索和筛选输入框。

清理 5.5 之前的 GPT 内置模型、供应商测试候选和网关旧兼容槽位，DeepSeek 改用当前型号槽位；历史日志保留原有价格计算。
Windows 构建提供包含宿主应用和资源的 x64-portable.zip 便携包；需要 WebView2，用户数据仍保存在正常的 Windows 用户目录。
Codex 代理显示样式在代理预览中选择，移除通用设置和快捷设置的重复入口。

Add official GPT-6.1 Sol cost estimates, including cached reads/writes, Fast/Flex rates and long-context pricing.
Add GPT-6.1 Sol to built-in model catalogs, wakeup/Pelican selection, provider test candidates and native gateway model recognition. Existing default model choices are preserved, and GPT-6.1 Sol supports the ultra reasoning effort in wakeup tasks and Pelican tests.
Model providers offer Pelican tests in the existing dialog, with saved provider/key/model selection, attributed results, shared history and retries. Tests use an isolated temporary gateway and resolve saved credentials again on retry.

Provider chat tests share the wakeup default model (gpt-5.6-luna) for the local Responses gateway and prefer it when available in the provider catalog. Explicit model choices and provider-specific upstream models remain supported.
Manual quota refresh now reports local proxy errors instead of failing without feedback.
Antigravity metadata scanning no longer probes WSL paths on Windows, so it cannot start WSL or stall while looking for a local client.
Pelican results stay on the batch you opened while other batches keep updating, and retrying a failed item resumes that batch instead of switching away.
Codex account toolbar actions no longer crowd the search and filter fields in narrow windows.

Remove pre-5.5 GPT models from built-in catalogs, provider test candidates and legacy gateway shells. DeepSeek mappings use current shells; historical logs retain their original pricing support.
Windows builds include an x64-portable.zip package with the host application and bundled resources; WebView2 is required and user data remains in the normal Windows profile.
Codex proxy display styles are selected from the proxy preview; duplicate general and quick-setting controls have been removed.