Features, not speed claims · Checked November 2026
How QECTOR Compares
PyMatching, ldpc and fusion-blossom are excellent open-source decoders. QECTOR's case is breadth: many decoder families, qLDPC and GPU batch decoding behind one API, with every correction checked.
| Feature | QECTOR Decoder v3 | PyMatching | ldpc | fusion-blossom |
|---|---|---|---|---|
| Main decoders | Union-Find, exact MWPM (Blossom, Sparse Blossom), BP-OSD, Relay-BP | MWPM (Sparse Blossom), correlated matching | BP+OSD, BP+LSD, BeliefFind (BP + Union-Find) | Parallel MWPM |
| Graph-like codes (surface, repetition) | Yes | Yes, its specialty | Yes | Yes |
| qLDPC and non-graph-like codes | Yes (BP-OSD, Relay-BP) | No: needs graph-like error models | Yes, its specialty | No |
| Several families on one check matrix | Yes, one API | Matching only | BP-based family | Matching only |
| GPU batch decoding | Yes (CUDA, OpenCL; paid plans) | No | No | No |
| Stim and Sinter | Yes | Yes | Via Sinter decoder classes | Via Sinter |
| Core language | Rust, Python API | C++, Python API | C++, Python API | Rust, Python API |
| Licence | PolyForm Noncommercial; commercial plans | Apache 2.0 | MIT | MIT |
| Commercial support | Yes, with offline-verified licences | Community | Community | Community |
Choose another tool when
- You only decode surface or repetition codes with matching and want the most established tool: PyMatching is the reference.
- You need a permissive open-source licence to redistribute the decoder inside your own product for free.
- You only need BP-based qLDPC decoding and are happy on CPU: ldpc is mature and MIT-licensed.
Choose QECTOR when
- You compare several decoder families on the same codes and want one API and one output format.
- You work on qLDPC codes and want GPU batch decoding.
- You need every correction checked against its syndrome, and commercial support with licences that work air-gapped.
- You use Claude: the QECTOR Claude plugin brings verified decoding into Claude Code and Claude Desktop.
About this comparison
Features and licences come from each project's own PyPI page, checked on 6 November 2026. We publish no speed comparison: decoder speed depends on code, noise model, hardware and settings, so measure on your own workload. QECTOR ships a PyMatching compatibility shim (from qector_decoder_v3.pymatching import Matching) so you can switch and compare in place. Spotted something out of date? Write to [email protected] and we will correct it.
Sources: PyMatching · ldpc · fusion-blossom · QECTOR Decoder v3