Inference Markets
Every frontier model through one endpoint, paid for in $COMD and burned.
Inference Markets sells the models themselves. Send $COMD, take an API key, and call every frontier model through one endpoint.
The $COMD you spend is burned. Buying inference takes supply off the market for good.
#The endpoint
https://api.comd.fun/icm/v1It speaks the standard chat-completions shape, so any client that takes a base URL and a key works unchanged. Put your key in the bearer and name the model you want.
curl https://api.comd.fun/icm/v1/chat/completions \
-H "Authorization: Bearer cicm_live_..." \
-d '{"model":"anthropic/claude-sonnet-4",
"messages":[{"role":"user","content":"hello"}]}'One endpoint reaches Claude, GPT, Gemini, Llama, DeepSeek, Mistral, Qwen, Grok and the rest. No card, no monthly bill, and no account of your own with a model provider.
#Paying per call
The rate from $COMD to credit is on the Inference Markets page and at GET /icm. A call holds an estimate when it opens and settles on the provider's own usage figure when it closes, so you are never charged for a call that did not happen, and an interrupted call releases what it held. Every answer states what it cost in usage.cost.
#Buying credit
- Pay in $COMD. One transfer to the desk.
- Pay in ETH. Swapped into $COMD on the way in, in a single transaction. This needs the $COMD pool, which opens when $COMD graduates on Pons; until then the tab says so.
- Send it yourself to the deposit address from any wallet.
The credit lands on the wallet that paid. Coming from another chain, bridge ETH to Robinhood Chain first: Relay and Across take seconds, and the canonical bridge is the slower trust-minimised route from Ethereum.
#Keys
One signature proves a wallet is yours and mints a key. The signature is not a transaction. The secret is shown once, because only a hash of it is stored. A key can carry a spend ceiling, and every key shows what it has spent and when it was last used. Revoking costs nothing.
#In the open
The desk's activity is public at GET /icm/v1/stats and live on the page: calls served, tokens processed, the most used models, median response time, and every $COMD paid in and burned. The page also has a playground to try any model with your key, and quickstarts for cURL, Python, TypeScript and agent frameworks.