Google has released the LiteRT Tensor API, a C++ layer in its LiteRT on-device machine learning stack that lets developers build a model's computation graph by hand, op by op, instead of converting a trained model from another framework. The company announced it in an Oct. 19, 2026 post on the Google Developers Blog .

Google isn't steering most developers away from conversion. Converting a trained model, for example a PyTorch model with LiteRT Torch, "offers the quickest and most reliable route to production, and it's the path we recommend for most projects," the post said.…