According to CoinMeta, OpenRouter has launched Jev and router, integrating the Jev of typesafe into the model routing system. Before each request is sent, Jev assesses the task difficulty, precision requirements, as well as whether the current model and inference settings are sufficient. Based on this assessment, it decides whether to continue using the same model, increase the inference intensity, or switch to a different model. OpenRouter's previous approach tended to select models based on task type, whereas Jev and router continue to evaluate the difficulty of the current task. For simpler tasks, the inference intensity can be reduced; for more challenging tasks, it may be increased. If the same model can still handle the task, there is no rush to switch models. However, in longer conversations, once a model is changed, the cached context usually cannot be reused directly, and the new model may need to process the entire conversation record from scratch. OpenRouter's self-tests show that among 4 tests, Jev and router completed 237 tasks with a resolution rate of about 82%. In another 5 tests, the median latency of Jev and router was lower than that of other routers.