SAME INTELLIGENCE TIER HALF THE BILL
radium.cloud — change one model string and compare the bill. The first $25 is on us, no card needed. 
digraph {
rankdir=TB;
node [shape=box style=rounded fontsize=13];
app [label="your app — untouched" color="#14532d"];
strings [label="change 3 strings" color="#7cb342"];
radium [label="radium endpoint" color="#14532d"];
bill [label="half the bill" color="#7cb342"];
app -> strings -> radium -> bill;
}
https://api.radium.cloud/v1 speaks OpenAI · https://api.radium.cloud speaks Anthropic — same prompts, same SDKs, same agents.
What one free $25 buys — three lanes:
| Lane | Model string | Rate in / out per MTok | The $25 covers |
|---|---|---|---|
| Volume | Tycho 1.0 |
$0.50 / $2.25 | 50 million input tokens |
| Copilots | Clarke 1.0 |
$1.50 / $7.00 | ~16M in · ~3.5M out |
| Deep work | Hal 1.0 |
$2.75 / $13.75 | ~9M in · ~1.8M out |
The headline 50 million lives in the volume lane — summarizing, extraction and routing run there all day. 
The real diff — paste-ready
- base_url = "https://api.openai.com/v1" model = "gpt-…"
- base_url = "https://api.anthropic.com" model = "claude-…"
+ base_url = "https://api.radium.cloud/v1" model = "Tycho 1.0" # OpenAI format
+ base_url = "https://api.radium.cloud" model = "Hal 1.0" # Anthropic format
LangChain, Vercel AI SDK, LiteLLM, plain curl — framework recipes · route 10% of live traffic first with the staged migration guide
Three kicks that stretch the credit:
Prompt caching −90% on repeated context — the same $25 runs way past 50M- Card stays in your wallet — homepage, pricing and tokenomics all say no card
Vendor-claimed savings — your own bill from the free $25 is the real proof
Run the comparison on their dime. 
!