Run BGE Reranker v2
Ready in about 90 seconds. No install, no setup. From $0.22 per hour.
What it does
A reranker takes the top results from a search and reads each one properly against the question, then reorders them.
Reranking is slow, so you do not run it on the whole archive. You search first, take the top fifty, and rerank those.
The improvement is large. In most systems this single step does more for answer quality than any change to the search itself.
What you can ask it
Fifty candidate documents for one customer question.
You get backThe same fifty, reordered so the actual answer is in the top three.
Search results across multiple languages.
You get backRanking that works across languages, not just within one.
A shortlist of near-identical product entries.
You get backThe one that genuinely matches, ranked first.
Your setup
We run this on an RTX 3090. It reranks about 400 documents per second against a query. $0.22 per hour, billed by the minute.
Why this card?
At 568 million parameters this needs about 6GB. Reranking is normally run live, inside a search request, where what matters is the delay on a single small batch rather than raw throughput. A cheap card handles that well, and the money saved is better spent on running it against more candidates.
- Memory
- 24GB
- Speed
- about 2 minutes per thousand queries
- Rate
- $0.22/hr
What will it cost you?
Billed by the minute on a RTX 3090 at $0.22 per hour. A batch job shuts the GPU off when the last thousand queries is done, so this is the whole cost.
Two ways to run it
Click and use it
Paste a question and a list of candidates, see the reordering and the scores.
Get early access Best for exploring.Hand us the whole pile
Rerank a stored set of search results in bulk to measure the improvement.
Join the list Best for volume. The GPU shuts off automatically when it is done.Model facts
| Parameters | 568M |
|---|---|
| License | Apache 2.0 |
| Memory required | 6GB minimum |
| Base model | XLM-RoBERTa large |
| Publisher | Beijing Academy of Artificial Intelligence |
| Recommended hardware | RTX 3090, 24GB |
| Hugging Face | BAAI/bge-reranker-v2-m3 |
Want BGE Reranker v2 the day it opens?
We are testing the rental flow with a small group first. Leave an email and we will tell you when BGE Reranker v2 is ready to run. One message, no newsletter.
BGE Reranker v2 questions
Do I need to install anything to use BGE Reranker v2?
No. We start a machine with BGE Reranker v2 already loaded and hand you a link. Everything runs in your browser, there is nothing to download, and nothing is left on your computer afterwards. It works the same on a Mac, a Windows laptop, or a Chromebook. A workspace is usually ready in about 90 seconds.
How much does it cost to run 50 thousand queries reranked?
About $0.38. BGE Reranker v2 takes about 2 minutes per thousand queries on the RTX 3090 we recommend, so 50 thousand queries reranked is roughly 1 hour 44 minutes of GPU time at $0.22 per hour. Billing is by the minute, and a batch job shuts the GPU off the moment the last item finishes. The calculator above works this out for your own numbers.
Can I run BGE Reranker v2 on a cheaper card?
Sometimes, and the calculator will not always make it look worth it. At 568 million parameters this needs about 6GB. Reranking is normally run live, inside a search request, where what matters is the delay on a single small batch rather than raw throughput. A cheap card handles that well, and the money saved is better spent on running it against more candidates. If you want to try a different card anyway, the advanced catalog lets you pick one and shows the estimated time before you commit.
Can I use BGE Reranker v2 commercially?
Yes. BGE Reranker v2 is released under the Apache 2.0 license, which permits commercial use with no conditions worth worrying about. You can use the output in paid client work and in products you sell.
What happens to my files after I stop?
Anything you save into your workspace folder stays in your account and is there when you come back. Everything else is destroyed: the machine is terminated and its disk is wiped. Batch job inputs and outputs are deleted from our storage 24 hours after you download them. We do not read your files or use them to train anything.
What if it fails partway through?
You are not charged for work that did not complete, and failed sessions are refunded automatically. Batch jobs write each output as it finishes rather than at the end, so a failure at item 380 of 400 leaves you 379 usable files and a message saying what broke. Restarting picks up where it stopped rather than redoing the work.
Own a GPU that sits idle? Put it to work. The average provider earns $180 to $420 per month per card.
List your GPU