GPT, Claude and Gemini from one API key
Send the same prompt to OpenAI, Anthropic, Google and hundreds of other models with one Superpowers key, through the REST API or MCP. Every model takes the same input (prompt or messages, max_tokens) and returns the same output (text plus token counts), so switching models changes one string. Models are routed through OpenRouter and billed per call at their token prices plus a small fee; failed calls are free.
Real example
Call made on 10/08/2026 through Claude Haiku 5.5 via OpenRouter, answered in 2506 ms. From a three-model comparison on the same prompt; Claude's answer shown.
curl -X POST https://superpowers.tools/v1/run \
-H 'Authorization: Bearer $SUPERPOWERS_KEY' \
-H 'content-type: application/json' \
-d '{"tool":"openrouter.anthropic-claude-haiku-5.5","idempotency_key":"example-1","input":{"prompt":"In two sentences, what is the Model Context Protocol and why does it matter for AI agents?","max_tokens":200}}'{
"text": "The Model Context Protocol (MCP) is an open standard, introduced by Anthropic in 2024, that defines how AI applications connect to external data sources, tools, and services through a common interface. It matters for AI agents because it replaces one-off, custom integrations with a single standardized way to give models access to real-world context and actions, making agents easier to build, extend, and interoperate across different tools and platforms.",
"model": "anthropic/claude-haiku-5.5",
"input_tokens": 37,
"output_tokens": 130
}Providers
Same input and output for every provider. Success rate over 30 days shows after 5 real calls.
| Provider | Price per call | Status |
|---|---|---|
| Claude Haiku 5.5 | $0.000378 | Available |
| GPT-5.4 Mini | $0.00324 | Available |
| Gemini 3.8 Flash | $0.002835 | Available |
| Claude Sonnet 5.5 | $0.00756 | Available |
| GPT-5.6 Sol | $0.00756 | Available |
| Gemini 3.1 Pro Preview | $0.00864 | Available |
| GPT-5 Nano | $0.00027 | Available |
| Gemini 2.5 Flash Lite | $0.000324 | Available |
| Nemotron 3 Super (free) | Free | Available |
| Nemotron 3 Ultra (free) | Free | Available |
| Lyria 3 Pro Preview | Free | Available |
| North Mini Code (free) | Free | Available |
Returned fields
text | The model's answer |
model | The model that answered, as named by OpenRouter |
input_tokens / output_tokens | Tokens billed; output includes any reasoning tokens |
input: prompt or messages | A single prompt, or a chat history with system and user messages |
input: max_tokens | Upper limit on output tokens, reasoning included |
Use it from your AI agent
Add the MCP server to Claude, Cursor or another client and ask in plain language.
claude mcp add --transport http superpowers https://superpowers.tools/mcp \ --header "Authorization: Bearer $SUPERPOWERS_KEY"
{
"mcpServers": {
"superpowers": {
"url": "https://superpowers.tools/mcp",
"headers": {
"Authorization": "Bearer ${env:SUPERPOWERS_KEY}"
}
}
}
}codex mcp add superpowers --url https://superpowers.tools/mcp \ --bearer-token-env-var SUPERPOWERS_KEY
Frequently asked questions
How much does a call cost?
Each model's token price from OpenRouter plus a small fee, charged per call. On 2026-10-08 the same short prompt cost $0.000075 on Claude Haiku 5.5, $0.00039 on GPT-5.4 mini and $0.00081 on Gemini 3.8 Flash. Each model page lists its price per million input and output tokens.
Which was fastest?
In that run GPT-5.4 mini answered in 1.8 s, Gemini 3.8 Flash in 2.1 s and Claude Haiku 5.5 in 2.5 s. That is one run; the compare page shows 30-day averages.
Why did one model return a cut-off answer?
Reasoning models spend part of max_tokens thinking before they write. With max_tokens 200, Gemini 3.8 Flash used 196 tokens and stopped after one phrase; with 1,000 it answered in full (513 output tokens, 5.3 s, $0.0021). Give reasoning models more room.
Can I compare models on my own prompt?
Yes. POST /v1/compare with the category llm.chat and a list of models runs them all on the same input and charges each at its own price; the three-model comparison above cost $0.0013. The compare page does the same in the browser and returns a share link.
How does my AI agent use it?
Add the Superpowers MCP server to Claude, Cursor, Codex or ChatGPT and the agent can call any model with run_tool, or call POST /v1/run with a tool such as openrouter.openai-gpt-5.4-mini. New accounts get $5 of credit.
Start with $5 of credit
Enough for 13,227 calls to Claude Haiku 5.5. No card needed to start.