Half Off Dev docs

Half Off Dev gives you the frontier models from Anthropic, OpenAI and Google for 50% of what their official price lists charge, on every model. The catch is that there isn't one: you keep your SDK, your request bodies and your favourite CLI, and change two values, the base URL (https://halfoff.dev) and the API key. Billing is per token from a prepaid balance.

Try it with the SDK you have#

Here is a Claude request through the Anthropic SDK. Compared with calling Anthropic directly, only the base URL and the key differ.

curl https://halfoff.dev/v1/messages \
  -H "x-api-key: $HALFOFF_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-sonnet-5.5",
    "max_tokens": 512,
    "messages": [{"role": "user", "content": "Give me three names for a code review bot."}]
  }'

The same key also calls GPT models through the OpenAI SDK and Gemini models through the Google Gen AI SDK.

Where to point each SDK#

Why it's a drop-in swap#

• Half the list price, every model. Each token costs 50% of the official rate for that model. Live prices are in GET /v1/models and on Pricing.
• No format translation. A Messages request is answered as a Messages request, a Chat Completions request as Chat Completions, a Gemini request as Gemini. Streaming, function calling, extended thinking and prompt caching behave as the vendor documents them. See How the gateway works.
• Three channel groups. Claude · Value, GPT · Value and Gemini · Value run on pooled capacity. Short bursts wait a few seconds in a queue rather than erroring, and a failed call is retried on another channel before it reaches you. See Channel groups and Rate limits.
• Caches stay warm. A conversation sticks to the channel holding its prompt cache, so repeated context keeps its cached-input price. See Prompt caching.
• Prepaid, no invoices. You top up a balance with PayPal and requests draw from it. At zero, requests stop; nothing is billed after the fact. See Billing.
• A log line per call. Tokens, charge, latency and status for each request land in your dashboard logs. See Usage.

Need a hand?#

For a question about a specific call, a charge or your account, open a ticket from Support and paste the x-halfoff-request-id response header if you have it. Live service health is on the status page.