Tengine review
Open-source C/C++ inference with CPU, GPU and NPU support for embedded deployment.
Reviewed by iTechGuides Editors · Editorial team · Updated Oct 2026
Tengine is an Apache-2.0 open-source neural-network inference engine for embedded and AIoT devices. It is aimed at teams deploying deep-learning models through C or C++ on Linux, Android, desktop, mobile and other embedded targets. The project provides C and C++ inference APIs, CMake-based cross-platform compilation, computational graph lifecycle execution and deployment options spanning Linux x86/ARM, Android and NPU hardware.
Its main strength is the separation between model handling and hardware execution. CPU, GPU and NPU backends can run through the same front-end/back-end architecture, while TIM-VX integration connects Tengine with NPU hardware. The published feature set covers model conversion to tmfile and support for Caffe, MXNet, ONNX, TensorFlow, TFLite, DarkNet, NCNN, MegEngine and PaddlePaddle formats. Quantization utilities for uint8 and int8 models, along with benchmark and speed-assessment tools, give engineering teams ways to prepare and measure deployments across heterogeneous hardware.
Tengine is a fit when an organization wants open-source, self-hosted inference and can own compilation, model preparation and device integration. CMake builds and low-level C/C++ APIs suit embedded engineering workflows, but they also place more setup responsibility on the adopting team than a managed inference service would. Documentation, community support and email are available support channels. Choose Tengine for portable embedded execution and hardware-backend control; teams seeking a hosted service or a higher-level application interface should consider an alternative designed around managed deployment.
Tengine pros and cons
- Where it wins
- Runs workloads on CPU, GPU and NPU backends
- Supports broad model formats with tmfile conversion
- Includes quantization, benchmarking and speed-assessment tools
- Where it doesn't
- Self-hosted deployment puts build and integration work on the team
- C/C++ and CMake workflows may require systems expertise
- Focused on embedded inference rather than a hosted service
Tengine fact sheet, pricing and score →
Advertiser disclosure: iTechGuides is reader-supported. Vendors can pay for top positions in our rankings and for a place on other products' pages, and we may earn a commission when you click some links. Bought places stand at the number bought; this page and our disclosure say so. Paying vendors may get a fuller, more detailed review from our editors, and that review can change their score. How we rank.
Last updated · How we research and update