Back to Gap Map Model components / Inference code

ncnn

open source / Overall score: 3.0

ncnn is a neural network inference runtime optimized for mobile, embedded and desktop deployment, with no third-party dependency in its core. It ships pnnx for converting PyTorch and ONNX models into its own format, and targets general neural network inference rather than LLM serving specifically.

Openness

5 medium confidence

5.0

license: BSD-3-Clause(OSI)
source: public(the published repository is the engine)
core features withheld: no — no enterprise path in the repository root and no paid build of the engine in the README

The LICENSE body is a Tencent notice licensing both the ncnn binary and source under BSD-3-Clause, followed by the separate terms of the third-party components it vendors. The repository is public, active and builds the engine itself, with no enterprise or paid tier in the root tree or README.

https://api.github.com/repos/Tencent/ncnn recorded 2026-08-31
Repo metadata for Tencent/ncnn - archived false, 23,768 stars.
https://raw.githubusercontent.com/Tencent/ncnn/master/LICENSE.txt recorded 2026-10-07
LICENSE body, read in full: a Tencent-authored notice stating that both the ncnn binary and the ncnn source are licensed under the BSD 3-Clause License, followed by the separate terms of third-party components it vendors.
https://raw.githubusercontent.com/Tencent/ncnn/master/README.md recorded 2026-10-07
README, read for a paid tier, an enterprise edition or a licence-gated build beside the published source, and for whether any component is stated to be private.

Adoption

3 low confidence

3.0

An ncnn package exists on PyPI and resolves to this repository, but pypistats publishes no download figure for it, so no package artifact can be declared. GitHub stars are the only signal left, and they cap the reading because a star is not a use.

https://api.github.com/repos/Tencent/ncnn recorded 2026-08-31
Repo metadata - stargazers_count = 23,768 - for Tencent/ncnn.

Capability

3 medium confidence

3.0

ncnn sits two tiers below vLLM. Its platform reach is wide, but vLLM is judged on LLM-serving features, batching, paged attention, parallelism and an OpenAI-compatible surface, none of which ncnn provides, which keeps it with the other narrow-purpose runtimes rather than with ONNX Runtime one tier up.

https://raw.githubusercontent.com/Tencent/ncnn/master/README.md recorded 2026-10-07
README, read for the engine's feature set: General neural network inference across mobile, embedded and desktop targets, with a model conversion toolchain; not an LLM serving engine.

Verified 2026-09-27