Unfragile

For cross-encoder re-ranking, the most direct fit is a hosted BGE reranker model on Replicate 1, which runs the BAAI bge-reranker-v2-m3 model via API 11. Sentence Transformers (SBERT) is the standard Python library for using and training reranker models 2, and its page lists no explicit re-ranking action record, so it is an alternative to self-hosting.

1
Replicatereplicate.comverified and signed
Hosted cloud API that runs open-source models, including a BGE M3 reranker (sesamo-srl/bge-reranker-v2-m3, the BAAI reranker v2-m3 with FP16 inference) that scores query-document pairs 111.
Website Docs Pricing Get started
for agents: sesamo-srl/bge-reranker-v2-m3
2
Sentence Transformers (SBERT)sbert.netverified and signed
Python library for using and training embedding and reranker models, usable for cross-encoder re-ranking locally 2.
Website Docs
official:SDK · sentence-transformers+24 community tools built for it
for agents: XLNet · embedding · Transformer
3
Hugging Face Hub client libraryhuggingface.coverified and signed
Python client (huggingface_hub) to discover and download pre-trained models, including reranker checkpoints, from the Hugging Face Hub 313.
Website Docs Pricing Get started
official:SDK · huggingface-hub
for agents: model-hub · models · pretrained-models

Sources

1Replicate - Run AI with an APIreplicate.com · read 2026-10-29
2SentenceTransformers Documentation — Sentence Transformers documentationsbert.net · read 2026-10-29
3🤗 Hub client library · Hugging Facehuggingface.co · read 2026-10-29
11sesamo-srl/bge-reranker-v2-m3: declares "sesamo-srl/bge-reranker-v2-m3" (the record we read and signed)unfragile.ai · read 2026-10-24
13huggingface-hub: declares "model-hub" (the record we read and signed)unfragile.ai · read 2026-10-24

For agents: the 10 actions behind this answer

sesamo-srl/bge-reranker-v2-m3 · sesamo-srl/bge-reranker-v2-m3 verified and signed
Newest reranker model from BAAI (https://huggingface.co/BAAI/bge-reranker-v2-m3). FP16 inference enabled. Normalize param available
yxzwayne/bge-reranker-v2-m3 · yxzwayne/bge-reranker-v2-m3 verified and signed
Newest balance-striking reranker model from BAAI. Outputs rank scores for query-doc pairs. FP16 inference enabled.
ninehills/bge-reranker-base · ninehills/bge-reranker-base verified and signed
BAAI/bge-reranker-base Model with fp16
tejasunku/qwen3-reranker-8b · tejasunku/qwen3-reranker-8b verified and signed
Rerank query search results using qwen3 reranker 8b model, with support for instructions
XLNet · sentence-transformers verified and signed
model-hub · huggingface-hub verified and signed
· async-std website read, tools not yet checked
A Rust library for easy and understandable async programming
yuval-alaluf/restyle_encoder · yuval-alaluf/restyle_encoder verified and signed
ReStyle: A Residual-Based StyleGAN Encoder via Iterative Refinement
· aio-libs Multidict website read, tools not yet checked
krthr/clip-embeddings · krthr/clip-embeddings verified and signed
Generate CLIP (clip-vit-large-patch14) text & image embeddings

Labels are the instruments' verdict and are never changed by ranking. Agents and robots get this same answer as JSON at /api/v1/find?q=…&answer=1 and through the MCP server.