Torch launches unified model API and MCP inference layer
Torch says its API connects 470 models across 59 labs, supports ensemble inference, and opens the inference layer to AI agents via MCP, with token payments for inference.
$ANTI & $TORCH IS BUILDING WHAT MOST AI TOKENS ONLY TALK ABOUT 🧠🔥 TORCH and ANTI are shipping real tech at a pace we haven’t seen in a long time. The execution speaks for itself, and what they’ve already achieved is just the beginning. This is what building in crypto is supposed to look like. Launched just days ago on Robinhood Chain through @Torchpaddotfun, and it’s already shipping a working product with token mechanics tied to actual usage. Here’s what caught my attention. The product is live: • One API connecting 470 models across 59 labs, including OpenAI, Google, Anthropic and Qwen, plus image models. • Ensemble inference lets users query multiple models at once. • The inference layer is now open to AI agents through MCP. • Holders get early model access, higher rate limits, price locks and ensemble features. But the token mechanics are what make this interesting. • $ANTI can be used to pay for inference. • 50% of daily realized margin goes toward buybacks. • The developer has burned 10% of the supply from their wallet. • Torch platform fees also feed into burns: 50% of the fee is used to buy and burn ANTI on the open market, with activity tracked on chain. The thesis is simple: real product usage + buybacks + burns. If inference demand grows, the model has a mechanism to connect that activity to the token. @tryantimatter × @Torchpaddotfun Route the models. Burn the supply. Let usage do the talking. cc- @log1_053 @coinesper @LordDurden @YorichiiCrypto