Claude · GPT · Gemini

Frontier models.Half the price.

Claude, GPT and Gemini through their native APIs at 50% off the official list price. Every model, every token, input, output and cache. Change one URL and keep your code.

Get your API key See every price

Native Anthropic, OpenAI & Gemini APIs
Prepaid, no subscription
Pay with PayPal

50%OFF

Receipt#26-0050

Claude Opus 5.5

A month of coding agent traffic

40M input$160
6M output$120
220M cache reads$44

Official price$324

You pay$162

You keep$162

Thank you for not paying full price.

Switching takes one line.

Same SDKs, same request bodies, same streaming, tools, thinking and prompt caching. Only the base URL and the key change.

Works with Claude Code, Codex CLI, Gemini CLI, Cursor, Cherry Studio and anything that speaks these APIs.

Setup guides

agent.py

from anthropic import Anthropic
client = Anthropic(
-    base_url="https://api.anthropic.com",
+    base_url="https://halfoff.dev",
+    api_key=os.environ["HALFOFF_API_KEY"],
)
client.messages.create(model="claude-sonnet-5.5", max_tokens=1024, messages=msgs)

The price board.

Live prices per 1M tokens, straight from the catalog. The official list price on the left, ours on the right.

All models and prices →

ModelOfficial priceHere

Claude Opus 5.5Anthropic$4 / $20$2 / $10$4 / $20−50%

Claude Sonnet 5.5Anthropic$2 / $10$1 / $5$2 / $10−50%

Claude Haiku 5.5Anthropic$0.1 / $0.5$0.05 / $0.25$0.1 / $0.5−50%

GPT-6.1 SolOpenAI$2 / $10$1 / $5$2 / $10−50%

GPT-6 AstraOpenAI$10 / $50$5 / $25$10 / $50−50%

Gemini 4 ArgonGoogle$2 / $10$1 / $5$2 / $10−50%

Gemini 3.8 FlashGoogle$0.75 / $3.75$0.375 / $1.88$0.75 / $3.75−50%

Gemini 3.1 ProGoogle$2 / $12$1 / $6$2 / $12−50%

USD per 1M input / output tokens

What would you keep?

Drag to your monthly LLM bill at official prices.

Your bill today

$1,200

You pay here

$600

kept per year

$7,200

Cheap, not cut-rate.

Untouched native APIs

Anthropic Messages, OpenAI Chat Completions and Responses, and the Gemini API, passed through as you send them.

Caching that keeps paying

Each conversation stays where its prompt cache lives, so long agent sessions keep getting cache rates, at half price too.

Failover built in

Pooled capacity behind every model. A slow or failing upstream is skipped before your request ever sees it.

Spend you control

Prepaid balance, per-key spend caps, rate limits, model allow-lists and a log line for every request with its exact cost.

Questions

Is it really half price on every model?

Yes. Every Claude, GPT and Gemini model costs 50% of its official list price per token: input, output, cache reads and cache writes. The price board above is live.

How can it cost half?

We buy capacity in bulk across many pooled accounts and pass the difference on. You get the same models through the same APIs; we keep a margin on volume.

Do I have to change my code?

No. Point your SDK or tool at our base URL and use your key. Request and response formats are the vendors’ own.

How do I pay?

Add prepaid credit with PayPal, from $5. There is no subscription and no minimum. Requests stop when your balance runs out, so there are no surprise bills.

What about rate limits?

Traffic is spread over pooled capacity and queued at peaks instead of rejected. You can set your own limits and spend caps per key.

Stop paying full price.

Create a key, add credit, change one line.

Start now

halfoff.dev