Coding agents resend the same system prompt and file context every turn. HalfRouter passes Claude prompt caching through natively, so those tokens come back as cheap cache reads — and you pay in USDT, no card.
Long system prompts and file context bill as ~0.1× cache reads instead of full-price input, turn after turn.
Drop the base URL into Cline, Roo, Aider, Continue or Cursor and pick a Claude model. Done.
Standard tier for the lowest price, Dedicated for pure first-party upstreams with cascading failover.
API Provider : OpenAI Compatible Base URL : https://halfrouter.vip/v1 API Key : YOUR_HALFROUTER_KEY Model ID : claude-sonnet-4.6
Agents resend a large, mostly-identical context every turn. With native prompt caching, the repeated tokens are billed as cheap cache reads instead of full-price input — the savings compound over a session.
Anything OpenAI-compatible: Cline, Roo Code, Aider, Continue.dev, Cursor, LibreChat, plus the OpenAI SDKs.
Sonnet 4.6 is a strong default for agentic coding; Opus 4.8 for the hardest tasks; Haiku for cheap, fast edits.
Top up in USDT — pay-as-you-go, no card and no KYC.
Sign up, grab a key, top up in USDT — your existing OpenAI-style code runs on GPT, Claude & Gemini at half the bill.