Starstruk Logo MarkStarstruk

v0.2.1

Wei-Shaw/sub2apiv0.2.1Sep 5, 2026by github-actions[bot]

AI Summary

This release introduces support for GPT-6 Astra and Codex Ultrafast service tiers, alongside enhanced gateway capabilities to improve compatibility and stability for OpenAI, Claude, Gemini, and image generation services. It also includes significant optimizations for memory usage, pricing controls, and fixes various bugs across different models.

Key Highlights

  • Added GPT-6 Astra and Codex Ultrafast service tiers.
  • Enhanced image generation by downloading upstream URLs and filling them back as `b64_json`.
  • Improved memory efficiency with immutable data for WebSocket replay.
  • Added pricing controls for Anthropic reasoning effort and hot-loading capability.
  • Comprehensive bug fixes for Claude billing, OpenAI model persistence, and chat fallbacks.

New Features

  • OpenAI: Support for GPT-6 Astra model sync and message hint caching
  • Codex: Support for Ultrafast service tier and fixed account model list
  • Claude: Support for configuring the externally displayed CLI version
  • Image Generation: Support for downloading upstream URLs and filling them back as `b64_json`
  • Usage Recording: Save upstream request identifiers and support custom response header names per account
  • Pricing: Support for Anthropic reasoning effort pricing control and hot-loading pricing files
  • Reasoning Mapping: Support for using `none` as a reasoning mapping source

Full Release Notes

> AI API Gateway Platform - 将 AI 订阅配额分发和管理

新增 GPT-6 Astra、Codex Ultrafast 服务层级及多项网关能力，提升 OpenAI、Claude、Gemini 和图片生成场景的兼容性与稳定性。

## 新增功能

- OpenAI：支持 GPT-6 Astra 模型及能力同步、消息提示缓存
- Codex：支持 Ultrafast 服务层级和固定账号模型清单
- Claude：支持配置对外展示的 CLI 版本
- 图片生成：支持将上游图片 URL 下载并回填为 `b64_json`
- 用量记录：保存上游请求标识，并支持按账户配置响应头名称
- 定价：支持 Anthropic reasoning effort 定价控制及定价文件热加载
- 推理映射：支持将 `none` 作为 reasoning 映射来源

## 优化改进

- 管理端账户列表支持紧凑响应，减少返回数据量
- 前端侧边栏支持在子路由激活时折叠分组
- 网关调度补充渠道模型限制的逐账号检查
- WebSocket 重放正文改为共享不可变数据，降低长会话内存开销
- 记忆被上游拒绝的加密内容，避免后续轮次重复恢复

## Bug 修复

- 修复 Claude 计费指纹与出站 User-Agent 不一致
- 修复 OpenAI 模型列表不可用时丢失已有能力
- 修复 OpenAI Astra 能力和续聊状态未持久化
- 修复图片 URL 回填时访问私网地址及内容类型校验问题
- 修复 WebSocket 多路径下未记录 cyber policy 失败
- 修复上游请求标识记录不完整
- 修复 Gemini 自定义原生模型列表未生效
- 修复聊天回退时已发现工具丢失
- 修复 OpenCode 会话转发范围错误
- 修复 GLM-5.3 thinking effort 参数规范化
- 修复自动化 heartbeat bootstrap 处理
- 修复支付宝订单回调延迟导致订单长期待支付
- 修复转发失败后会话槽未及时释放
- 修复不可用 continuation 响应恢复
- 修复 API Key 透传时 reasoning effort 丢失
- 修复历史任务委派续聊兼容性
- 修复负载感知调度中的模型映射账号选择
- 修复上游错误代理归因和网关准入拒绝识别



---

## 📥 Installation

**Docker:**
```bash
# Docker Hub
docker pull weishaw/sub2api:0.2.1

# GitHub Container Registry
docker pull ghcr.io/wei-shaw/sub2api:0.2.1
```

**One-line install (Linux):**
```bash
curl -sSL https://raw.githubusercontent.com/Wei-Shaw/sub2api/main/deploy/install.sh | sudo bash
```

**Manual download:**
Download the appropriate archive for your platform from the assets below.

## 📚 Documentation

- [GitHub Repository](https://github.com/Wei-Shaw/sub2api)
- [Installation Guide](https://github.com/Wei-Shaw/sub2api/blob/main/deploy/README.md)