Head to head
Every platform on this site was measured in the same run, so any two of them can be compared without adjusting for hardware, documents or embedding model. These pages do that for the pairings buyers actually weigh.
How to read these
Each page opens with the answer, then shows the measurements behind it and computes which platform won each row from the data rather than asserting it. Where a figure is unreliable — a latency number distorted by a background queue, a rate covering only part of the corpus — the page says so instead of quietly using it.
Dify vs RAGFlow
Both ingested the entire corpus. Dify retrieved half again as many answers on a quarter of the memory; RAGFlow is the only one of the two you may legally resell.
FastGPT vs Dify
Both deployed on the same host against the same documents with the same embedding model. FastGPT preserves figures better; Dify finds them better, ingests everything, and does it eight times faster.
FastGPT vs RAGFlow
Both deployed on the same host against the same documents with the same embedding model. FastGPT keeps more of what it parses; RAGFlow ingests everything and does it in a quarter of the time.
A pairing you will not find here
Flowise appears on the ranking with deployment figures only. Its current published image does not start, and the open-source build offers no way to obtain an API credential without a browser, so there is nothing to compare it against on retrieval or parsing. The platform page sets out both findings.