New RunInfra

RunInfra is a model inference API developed by the YC-accelerated team RightNow, specifically designed to host open-source large models such as Kimi K3, DeepSeek V4.1 Flash, and GLM-5.3. A single API Key is compatible with both OpenAI and Anthropic SDKs; therefore, existing code can be easily modified by simply changing the connection address to use this API.

Its main selling point is “inexpensive caching”. During each round of dialogue, the agent sends large amounts of contextual information again and again. RunInfra ensures that the same session is always routed to the same hot copy. According to the official claims, the cache hit rate can reach 98%, and the cost for caching data is as low as 0.01 US dollars per million tokens. These optimization measures are implemented all the way down to the GPU kernel level.

In addition, programming packages have also been introduced. These packages can serve as the backend for more than 20 types of intelligent programming agents, such as Claude Code and opencode. The Starter package costs $10 per month, and its usage quota is reset every 5 hours and weekly. It is ideal for developers who wish to run their own intelligent agents and are sensitive to token billing expenses.