AI payment infrastructure Kite launches Marathon, an adaptive inference infrastructure for long-running AI Agents

On July 22, AI payment infrastructure Kite’s ecosystem launched Marathon, an adaptive inference infrastructure for long-running AI Agents. Users can choose a completion window for each request (now / soon / later / anytime—four tiers); the longer the wait, the lower the price. The deepest tier can save up to about 65% versus the real-time price (the discount is based on a reference value and fluctuates with real-time capacity). The service API is compatible with OpenAI and provides one-line install command plugins for Claude Code and Codex. At launch, it supports five top open-weight models: Kimi K3. GLM 5.2, DeepSeek V4 Pro, Qwen3.6-35B-A3B, and Nemotron 3 Ultra—all of which follow the same pricing rules. For details, see the official website marathon.build. It is learned that Kite previously completed a $33 million funding round led by PayPal Ventures and General Catalyst. This launch of Marathon is a key move as it extends from Agent identity and the payment network into the underlying compute infrastructure of Agent economics, aiming to connect the complete loop of “autonomous calling and autonomous payment” for Agents. #Ai