v1.12.5

This release improves usage accuracy, model pricing behavior, and pricing setup reliability. It also includes a small settings UI polish for scheduled quota refresh.

What's Changed:

• Improved token normalization for OpenAI-compatible providers with Gemini-style reasoning tokens.
• Kept strict OpenAI/Codex token handling separate from OpenAI-compatible provider handling.
• Updated model pricing to prefer real model names, using model aliases only as a fallback when no real-model price is configured.
• Improved pricing model options by merging CPA models with locally observed usage models, while keeping local models available if CPA model listing is unavailable.
• Fixed the scheduled refresh settings layout so frequency controls remain readable and stable.

---

本版本改进了用量统计准确性、模型计价行为和价格设置可靠性，并包含定时 quota 刷新设置界面的细节优化。

更新内容：

• 改进 OpenAI-compatible provider 中 Gemini 风格 reasoning tokens 的 token 归一化。
• 将严格 OpenAI/Codex token 处理与 OpenAI-compatible provider 处理分开，避免互相影响。
• 模型计价改为优先使用真实模型名，只有在真实模型未配置价格时才回退到模型 alias。
• 模型价格设置会合并 CPA 模型列表和本地已使用模型；当 CPA 模型列表不可用时，仍保留本地模型。
• 修复定时刷新设置布局，使刷新频率控件更稳定、更易读。