Java LLM integration compared: SDK, HTTP, Spring AI and LangChain4j

On November 10, 2026, a post on Juejin put four ways of wiring an LLM into a Java application to the test: the official SDK, plain HTTP, Spring AI, and LangChain4j. The author wrote four separate implementations that all ask the same question, who are you, so the four approaches can be read side by side. Alongside the code, the post includes a decision diagram for picking an approach and a cost comparison of multi-turn conversations, so developers weighing a Java-side integration can check which option fits their needs. It also rounds up 5 version pitfalls, namely issues tied to specific library versions that the author ran into during the experiments. For teams that are still deciding how to connect a large language model from Java, the piece is meant as a practical reference they can compare against their own setup. At present only this one source is available, and it does not disclose the underlying cost figures or the details of those version pitfalls, so the comparison stops at a high-level recommendation rather than a final verdict.