Get API Key → 获取密钥 →
Developer Documentation开发者极速接入指南

Connect Your Coding CLI in 60 Seconds 60 秒接入三大终端 AI 编程神器

One LightAI API key (sk-...) unlocks native Claude 5.5 / Fable 5.2 and GPT-6 / 6.1 across Claude Code, OpenCode, and Codex CLI at a flat 70% discount. 使用同一个 LightAI 密钥（sk-...），即可在 Claude Code、OpenCode 与 Codex CLI 中以官方 3 折直连最新一代 Claude 5.5 / Fable 5.2 与 GPT-6 / 6.1 旗舰模型。

Client客户端工具 Exact Base URL填写的 Base URL Trailing /v1?结尾是否带 /v1 Native Wire Protocol底层原生协议
Claude Code https://api.lightai.io NO /v1不带 /v1 Anthropic /v1/messages
OpenCode https://api.lightai.io/v1 Include /v1带 /v1 @ai-sdk/anthropic + openai
Codex CLI https://api.lightai.io/v1 Include /v1带 /v1 OpenAI /v1/responses

1. Claude Code

Claude Code speaks native Anthropic Messages (/v1/messages) with full Extended Thinking and 0.025×–0.05× Prompt Caching support. Claude Code 走原生 Anthropic Messages（/v1/messages）协议，满血支持 Extended Thinking 深度思考与低至 0.025×–0.05× 的提示词缓存省钱。

Important: Keep ANTHROPIC_BASE_URL as https://api.lightai.io (without /v1), because Claude Code appends /v1/messages automatically. 防踩坑提示：ANTHROPIC_BASE_URL 请填 https://api.lightai.io（结尾不要加 /v1），因为 Claude Code 会自动拼接 /v1/messages。

Option A · Quick Shell Launch方式一：终端环境变量快速启动

bash / zsh 
export ANTHROPIC_BASE_URL="https://api.lightai.io"
export ANTHROPIC_AUTH_TOKEN="sk-your-lightai-key"
claude

Option B · Persistent Config (Recommended)方式二：写入永久配置文件（推荐，一劳永逸）

Save to ~/.claude/settings.json so every terminal tab and IDE session automatically uses LightAI and defaults to claude-opus-5-5: 写入 ~/.claude/settings.json 后，所有新开终端窗口与 IDE 内置终端都会自动生效，无需每次 export：

~/.claude/settings.json 
{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.lightai.io",
    "ANTHROPIC_AUTH_TOKEN": "sk-your-lightai-key"
  },
  "model": "claude-opus-5-5"
}

2. OpenCode

OpenCode lets you switch freely between Claude 5.5 / Fable 5.2 and GPT-6 / 6.1 inside a single coding session. Our template configures @ai-sdk/anthropic for Claude (so native 0.025×–0.05× Prompt Caching triggers automatically!) and @ai-sdk/openai-compatible for GPT-6, with LightAI's exact 70%-off rates shown in OpenCode's live cost bar. OpenCode 支持在同一编程会话中自由切换 Claude 5.5 / Fable 5.2 与 GPT-6 / 6.1。下方配置为 Claude 启用原生 @ai-sdk/anthropic 驱动（自动激活低至 0.025×–0.05× 缓存省钱与思考块），并内置了全部 6 款纯血旗舰的精确 3 折单价，让 OpenCode 右下角实时费用统计与实际扣费分毫不差。

Option A · 1-Line Auto Installer (Recommended)方式一：一行命令自动配齐全部 6 款主力旗舰（推荐）

Automatically backs up any existing opencode.json to opencode.json.bak and merges all 6 LightAI flagship models: 自动备份你原有的 opencode.json 为 .bak，无损注入 LightAI 全系 Claude 5.5 + GPT-6.1 旗舰模型：

bash / zsh 
export LIGHTAI_API_KEY="sk-your-lightai-key"
curl -fsSL https://api.lightai.io/static/opencode.sh | sh
opencode

Option B · Full Manual Template (~/.config/opencode/opencode.json)方式二：手动复制完整模板 (~/.config/opencode/opencode.json)

~/.config/opencode/opencode.json 
{
  "$schema": "https://opencode.ai/config.json",
  "model": "lightai-claude/claude-opus-5-5",
  "provider": {
    "lightai-claude": {
      "name": "LightAI (Claude)",
      "npm": "@ai-sdk/anthropic",
      "options": {
        "baseURL": "https://api.lightai.io/v1",
        "apiKey": "sk-your-lightai-key"
      },
      "models": {
        "claude-fable-5-2": {
          "name": "Claude Fable 5.2",
          "reasoning": true,
          "tool_call": true,
          "attachment": true,
          "temperature": true,
          "cost": { "input": 2.40, "output": 12.00, "cache_read": 0.06, "cache_write": 3.00 },
          "limit": { "context": 200000, "output": 64000 }
        },
        "claude-opus-5-5": {
          "name": "Claude Opus 5.5",
          "reasoning": true,
          "tool_call": true,
          "attachment": true,
          "temperature": true,
          "cost": { "input": 1.20, "output": 6.00, "cache_read": 0.06, "cache_write": 1.50 },
          "limit": { "context": 200000, "output": 64000 }
        },
        "claude-sonnet-5-5": {
          "name": "Claude Sonnet 5.5",
          "reasoning": true,
          "tool_call": true,
          "attachment": true,
          "temperature": true,
          "cost": { "input": 0.60, "output": 3.00, "cache_read": 0.03, "cache_write": 0.75 },
          "limit": { "context": 200000, "output": 64000 }
        }
      }
    },
    "lightai-openai": {
      "name": "LightAI (OpenAI)",
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "baseURL": "https://api.lightai.io/v1",
        "apiKey": "sk-your-lightai-key"
      },
      "models": {
        "gpt-6-astra": {
          "name": "GPT-6 Astra",
          "reasoning": true,
          "tool_call": true,
          "attachment": true,
          "temperature": true,
          "cost": { "input": 3.00, "output": 15.00, "cache_read": 0.30 },
          "limit": { "context": 400000, "output": 128000 }
        },
        "gpt-6.1-sol": {
          "name": "GPT-6.1 Sol",
          "reasoning": true,
          "tool_call": true,
          "attachment": true,
          "temperature": true,
          "cost": { "input": 0.60, "output": 3.00, "cache_read": 0.03 },
          "limit": { "context": 400000, "output": 128000 }
        },
        "gpt-6.1-terra": {
          "name": "GPT-6.1 Terra",
          "reasoning": true,
          "tool_call": true,
          "attachment": true,
          "temperature": true,
          "cost": { "input": 0.30, "output": 1.50, "cache_read": 0.015 },
          "limit": { "context": 400000, "output": 64000 }
        }
      }
    }
  }
}

3. Codex CLI

Codex CLI and the Codex desktop app share ~/.codex/config.toml and communicate over the native OpenAI Responses API (/v1/responses). You can set model to gpt-6.1-sol, gpt-6-astra, or gpt-6.1-terra. Codex CLI 与 Codex 桌面端共用 ~/.codex/config.toml，底层走原生 OpenAI Responses 协议（/v1/responses）。主力日常编程推荐 gpt-6.1-sol，复杂架构攻坚可切换为 gpt-6-astra。

~/.codex/config.toml 
model = "gpt-6.1-sol"
model_provider = "lightai"

[model_providers.lightai]
name = "LightAI"
base_url = "https://api.lightai.io/v1"
experimental_bearer_token = "sk-your-lightai-key"
wire_api = "responses"
Tip: Put only your raw sk-... key inside experimental_bearer_token (do not add a Bearer prefix), then run codex directly in your project folder. 提示：experimental_bearer_token 内直接填纯 sk-... 密钥（不要加 Bearer 前缀），保存后在项目目录直接运行 codex 即可。

Models & Legacy Auto-Upgrade旗舰模型价格与老模型号自动升级

Many CLI tools hardcode older 4.x or 5.x model names for background title generation or sub-agents. LightAI silently maps all legacy aliases to their newest-generation flagship counterparts so requests never fail with "model not found": 许多客户端子任务或插件内部写死了旧版 4.x / 5.x 模型名。LightAI 网关在底层做了无缝兼容映射，无需修改客户端源码即可自动路由至最新一代旗舰：

Flagship Model ID最新旗舰模型号 Input / Output (1M)输入 / 输出 (每百万) Cache Read (30% of Official)缓存读取 (官方 3 折) Auto-Upgraded Legacy Aliases后台自动兼容映射的旧模型号
claude-fable-5-2 $2.40 / $12.00 $0.06 (0.025×) claude-fable-5, 5-0, 5-1
claude-opus-5-5 $1.20 / $6.00 $0.06 (0.05×) claude-opus-5 / 5-1 / 5-2, 4-8 / 4-7 / 4-6 / 4-5
claude-sonnet-5-5 $0.60 / $3.00 $0.03 (0.05×) claude-sonnet-5 / 5-1 / 5-2, 4-6 / 4-5
gpt-6-astra $3.00 / $15.00 $0.30 (0.10×) gpt-6
gpt-6.1-sol $0.60 / $3.00 $0.03 (0.05×) gpt-6-sol, gpt-5.6-sol, gpt-5.4 / 5.2 / 5.1 / 5
gpt-6.1-terra $0.30 / $1.50 $0.015 (0.05×) gpt-6-luna, gpt-5.6-terra / luna, gpt-5.1-mini / 5-mini

Prompt Caching & USDT Top-Up FAQ缓存计费与 USDT/USDC 充值说明

How does Prompt Cache Read billing save me money?Prompt 缓存读取计费是怎么帮我省钱的？

When using Claude Code, OpenCode, or Codex CLI on a large codebase, 85%+ of each turn repeats your system prompt and repository context. LightAI mirrors official cache multipliers at flat 30% of official rates: 0.025× on Fable 5.2 ($0.06 / 1M), 0.05× on Opus 5.5 ($0.06 / 1M), Sonnet 5.5 & GPT-6.1 Sol ($0.03 / 1M), and GPT-6.1 Terra ($0.015 / 1M), plus 0.10× on GPT-6 Astra ($0.30 / 1M). Cache creation is 1.25× once. No manual setup required! 在用 Claude Code、OpenCode 或 Codex 改项目代码时，每一轮对话有 85% 以上的上下文（项目文件与系统提示词）是重复的。LightAI 严格同步官方最新缓存读取倍率并同样按官方 3 折计费：Fable 5.2 为 0.025×（仅 $0.06 / 1M），Opus 5.5 为 0.05×（$0.06 / 1M），Sonnet 5.5 与 GPT-6.1 Sol 为 0.05×（$0.03 / 1M），GPT-6.1 Terra 仅 $0.015 / 1M，GPT-6 Astra 为 0.10×（$0.30 / 1M）；首次写入缓存为 1.25×。三大工具均已内置自动触发，无需任何额外配置。

How do USDT / USDC top-ups credit instantly?USDT / USDC 链上充值有什么注意事项？多久到账？

We support 1:1 USD top-ups from $1 via TRC20, BSC (BEP20), Arbitrum, Base, Polygon, and ERC20. Always transfer the exact decimal amount displayed on the checkout page (e.g. 10.02 USDT) on the matching chain — our on-chain listener matches orders by exact amount and credits your account automatically within 5–30 seconds. Balances never expire. 支持 $1 起充，1 USDT/USDC = $1.00 额度，覆盖 TRC20、BSC (BEP20)、Arbitrum、Base、Polygon 与 ERC20。转账时请务必核对所选公链，并严格按照收银台显示的精确小数金额转入（如显示 10.02 就转 10.02，交易所提币请注意扣除手续费后的实际到账金额），链上确认后 5～30 秒全自动秒上分，余额永不过期。

Common HTTP Error Codes常见报错状态码速查

Code Meaning & Fix原因与排查方法
401 Invalid or missing API key. Verify your sk-... token in Console → API Keys and ensure no extra spaces or Bearer duplication.API Key 无效或未传。请前往「控制台 → 令牌管理」核对 sk-... 密钥是否复制完整、有无多余空格。
402 / 403 Insufficient account or token quota. Top up in Console → Wallet or check if you set a spend limit on that specific token.账户余额不足或该令牌设置了额度上限。请前往「钱包充值」补足额度，或将令牌设为无限额度。
404 Wrong endpoint path — usually caused by adding /v1 to Claude Code's ANTHROPIC_BASE_URL or omitting /v1 in OpenCode / Codex CLI.接口路径错误——常见于 Claude Code 误加了 /v1，或 OpenCode / Codex 漏填了 /v1，请对照页面顶部表格修正。
429 Burst concurrency limit reached. CLI clients will automatically back off and retry within 1–2 seconds.瞬时并发过高触发限流保护，Claude Code / OpenCode 会自动退避并在 1～2 秒内重试成功。
502 / 504 Temporary upstream congestion or long reasoning timeout. Retry the turn; no quota is deducted on failed upstream requests.上游瞬时拥塞或超长思考超时。直接重试即可，未完成的失败请求不会扣除额度。