v0.2.11
getarcaneapp/arcanev0.2.11Sep 30, 2026by github-actions[bot]
Upgrading?
AI Summary
Adds new model support (GPT-6.1 Sol), Claude limit reset features, and introduces billing reservation and API key rate limiting.
Key Highlights
- Support for GPT-6.1 Sol model added
- Claude native limit reset redemption feature
- In-flight balance reservation to prevent overdraft
- API Key creation limits enabled (200 active, 60/hour)
Breaking Changes
- API Key creation limits enabled by default (configurable via api_key_create settings)
- In-flight balance reservation enabled by default (configurable via billing.inflight_reservation)
New Features
- GPT-6.1 Sol model support
- Claude native limit reset redemption
- In-flight balance reservation
- API Key creation limits
Full Release Notes
> AI API Gateway Platform - 将 AI 订阅配额分发和管理 新增 GPT-6.1 Sol 模型支持和 Claude 原生限额重置兑换;余额模式新增在途额度预占,防止并发请求导致透支。 ## 新增功能 - 支持 GPT-6.1 Sol 模型(含定价、模型目录与各兼容入口转换) - Claude 账号支持兑换原生限额重置:查询到可兑换额度后,可在账号列表中经二次确认一键重置 - 使用密钥弹窗的 Codex 配置支持远程模型目录(Codex 0.156.0+),旧版客户端仍可选择本地文件模式 - 识别 ChatGPT 最新订阅套餐类型,账号套餐标签显示更准确 - 新增 API Key 创建数量与频率限制:每用户最多 200 个有效 Key、每小时最多创建 60 次(可配置,0 表示不限制) ## 优化改进 - 支持 Astra Ultrafast 模型的能力识别与计费 - 开启「仅限 Claude Code」并配置了降级分组的分组,其 OpenAI 兼容入口(Chat Completions / Responses)改走降级分组,不再直接返回 403 ## Bug 修复 - 修复余额模式下多个高成本请求并发时可能透支余额的问题:请求准入时按预估成本预占在途额度,计费完成后释放 ## 破坏性变更 - API Key 创建限制默认开启:单用户有效 Key 超过 200 个或一小时内创建超过 60 次时,将无法继续创建。如需调整,修改配置 `api_key_create.max_active_per_user` / `api_key_create.max_per_user_per_hour`(设为 0 表示不限制) - 余额在途预占默认开启:余额较低的用户在并发请求时,可能比以前更早被拒绝。可通过 `billing.inflight_reservation` 配置关闭或调整 --- ## 📥 Installation **Docker:** ```bash # Docker Hub docker pull weishaw/sub2api:0.2.11 # GitHub Container Registry docker pull ghcr.io/wei-shaw/sub2api:0.2.11 ``` **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)
Related Repositories
ollama/ollama★ 184,247
Get up and running with Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.
fatedier/frp★ 110,069
A fast reverse proxy to help you expose a local server behind a NAT or firewall to the internet.
infiniflow/ragflow★ 92,635
RAGFlow is a leading open-source Retrieval-Augmented Generation (RAG) engine that fuses cutting-edge RAG with Agent capabilities to create a superior context layer for LLMs
gohugoio/hugo★ 90,198
The world’s fastest framework for building websites.
junegunn/fzf★ 83,715
:cherry_blossom: A command-line fuzzy finder