v2.9.34

dograh-hq/dograhv2.9.34Jul 4, 2026by github-actions[bot]

AI Summary

Introduces rate limiting cooldown mechanisms, cost statistics, and fixes for single model display and DeepSeek thinking requests.

Key Highlights

  • Rate limiting cooldown mechanism on 5xx errors
  • Temporary overload detection and channel cooldown
  • Fixed single model channel cost display
  • Fixed rate limit window unit semantics
  • Fixed DeepSeek Claude compatibility thinking requests

New Features

  • Automatic rate limiting on 5xx errors
  • Overload detection mechanism
  • Cost statistics display (USD)

Full Release Notes

### ✨ 新功能

- 上游返回 5xx 或 Retry-After 时触发限速冷却机制
- 新增上游临时过载检测与渠道冷却机制,提升故障转移稳定性
- 新增 `mdiCash` 前端图标

### 🐛 修复

- **修复单模型渠道费用统计不显示** — 修复 CostUSD 在单模型渠道下不展示的问题
- **修复限速窗口单位语义** — 限速窗口字段值现在直接以秒为单位,语义更清晰
- **修复 DeepSeek Claude 兼容端点 thinking 请求 400** — 过滤 DeepSeek 不支持的 context management 特性,修复 Claude Code 首轮/recap 请求的 thinking 回传被拒问题(来自 v2.9.33)

---

**Full Changelog**: https://github.com/BenedictKing/ccx/compare/v2.9.32...v2.9.34