Claude API for every Claude model, on one prepaid key

1. Create a key

   Sign up, add credit from $5, and create a key in API keys. You can give the key a spend limit, a model allow-list or a fixed channel group.

2. Point the SDK at Lovely Router

   Set the base URL to https://lovelyrouter.com. The SDKs also read ANTHROPIC_BASE_URL and ANTHROPIC_API_KEY from the environment, so code that builds a default client needs no edits.

3. Send a Messages request

   POST /v1/messages takes the body Anthropic documents. It reaches the model unchanged apart from the model id, and the response comes back byte for byte.

   cURLPythonNode.js

   Shell

   curl https://lovelyrouter.com/v1/messages \
     -H "x-api-key: $LOVELY_API_KEY" \
     -H "anthropic-version: 2023-06-01" \
     -H "content-type: application/json" \
     -d '{
       "model": "claude-sonnet-5.5",
       "max_tokens": 1024,
       "messages": [{"role": "user", "content": "Explain HTTP caching in two sentences."}]
     }'

The key works in x-api-key, which the Anthropic SDK sends, or in Authorization: Bearer. anthropic-version and anthropic-beta are forwarded as you send them.

Live prices in US dollars per million tokens, the same numbers GET /v1/models returns without a key. Spellings other tools use resolve to the same model: claude-sonnet-5-5, anthropic/claude-sonnet-5.5 and dated ids such as claude-haiku-4-5-20251001. The model marketplace shows live availability for each model, and status shows hour-by-hour availability per channel group.

Message Batches, Files and the retrieve-a-model endpoint are not served. Errors keep Anthropic's shape.

A prompt cache lives on the channel that wrote it. If each turn of a conversation landed on a different channel, each turn would pay to write the cache again. Lovely Router recognizes the turns of one conversation and sends them back to the channel that holds the cache.

You don't have to do anything: the fingerprint is the system prompt plus the first user message, which stay the same as a conversation grows. Claude Code's x-claude-code-session-id header, metadata.user_id or an x-session-id header take priority when present. A pin lasts up to an hour while the conversation is active. If its channel is busy, the turn waits briefly rather than writing a new cache; if the channel fails, the pin moves with the turn. Prompt caching shows how to check cache hits in your logs.

Claude Code speaks the Messages API, so it needs the base URL and your key:

export ANTHROPIC_BASE_URL=https://lovelyrouter.com
export ANTHROPIC_AUTH_TOKEN=$LOVELY_API_KEY
export ANTHROPIC_MODEL=claude-sonnet-5.5
claude

To keep these across sessions, put them under env in ~/.claude/settings.json, and set ANTHROPIC_DEFAULT_HAIKU_MODEL for quick background tasks. Every session stays on the channel that holds its cache. Use with your tools has the full setup, and AI coding agents covers running Claude Code, Codex CLI and Gemini CLI on one balance.

Claude is billed per token: uncached input, output (thinking included), cache reads and cache writes, each at the model's price per million tokens times the channel group's multiplier. Every group bills at list price today, so each Claude model costs what Anthropic lists, cache reads and writes included. The charge comes from the usage in the response and is deducted from your prepaid balance within seconds.

There is no subscription or monthly fee. Requests refused before they reach a model cost nothing, and when the gateway retries a failed call on another channel you pay only for the attempt that answered. A stream you close early pays for what was produced up to that point. Pricing lists every product on one page.