Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

mxbai-rerank-v3.1-listwise

Mixedbread's closed listwise reranker that reads the whole shortlist at once and follows plain-English ranking instructions.

Reviewed by ReviewRamble editors · Updated

Share this page:XLinkedInFacebookRedditWhatsAppEmail

Our rating

3.8/5

A listwise reranker that reasons about which document wins, at a fraction of an LLM's latency, as long as you are willing to keep your documents in Mixedbread's search.

What is mxbai-rerank-v3.1-listwise?

mxbai-rerank-v3.1-listwise is Mixedbread's current reranker, released on 27 July 2026 as a faster successor to the listwise v3 model from May. Most rerankers score each document on its own and sort the scores. A listwise model reads the whole candidate set together, so it can tell when a newer document supersedes an older one, at the price of more work per query.

It also takes natural-language instructions such as preferring recent documents or internal sources, which Mixedbread says it handles well on recency, source-priority and multi-step requests. Unlike v3, whose scores followed a fixed ladder by rank, v3.1's scores depend on the content, so a threshold means something.

The weights are not released and there is no standalone rerank endpoint. The model runs only inside Mixedbread Search, on documents uploaded to a Mixedbread store, where it is the default reranker. Mixedbread's open-weight mxbai-rerank-large-v2 remains available there as a pointwise option.

How much does mxbai-rerank-v3.1-listwise cost?

PlanPriceWhat you get
StarterFree$5 one-time credit, 100 queries a minute
Scale$20/month$20 of monthly credit, 2,400 queries a minute
Search with reranking$7.50 per 1,000 queries$4 semantic search plus $3.50 reranking
EnterpriseCustomVolume discounts, bring your own bucket or cloud

Billed as part of Mixedbread Search: semantic search $4 per 1,000 queries, plus $3.50 with reranking ($7.50 total). Indexing from $0.25 and storage $0.10 per million tokens a month. Free Starter with $5 one-time credit; Scale $20/month in credits.

Is mxbai-rerank-v3.1-listwise worth it?

The model is good at a job most rerankers ignore: deciding which of several relevant documents is the one that counts. Pointwise rerankers judge each passage alone, so a cancelled booking and its replacement both look relevant. Reading the shortlist together, v3.1 can follow a reschedule, an amendment or a revised forecast to the version in force, and Mixedbread's examples, from email to contracts, are all of that kind.

Mixedbread's figures are strong but entirely its own. On sampled ViDoRe v3 queries with Wholembed v3 as the first stage, v3.1 lifted NDCG@10 from 0.62 to 0.71 at 0.65 seconds a query. Cohere's Rerank 4 Pro reached 0.66 in 1.6 seconds, and OpenAI's gpt-5.6-sol at high effort reached 0.70 in about 40 seconds, the basis of the 61-times claim. The v3 model before it scored an MRR of 0.93 on a 900-example instruction test against 0.84 for Voyage rerank-2.5. No independent evaluation has appeared.

Speed improved over v3 without a new model size being disclosed. Median production latency fell by 25% for inputs under 8,000 tokens and 43% between 32,000 and 64,000 tokens, which Mixedbread puts down to a rewritten inference engine. Scores are now tied to content rather than position, so a cut-off can finally drop irrelevant results.

The constraint is the product around it. There is no rerank endpoint for results from your own search engine; documents must be uploaded to a Mixedbread store and searched there, at $7.50 per 1,000 queries with reranking. That price buys retrieval too, but next to Cohere Rerank 5 Pro at $0.05 per million tokens it is a different kind of commitment. Teams that want the listwise behaviour and can move their corpus will find it here; teams with an established search stack cannot bolt it on.

Where it earns its place

  • A legal-tech developer indexing contracts and their amendments

    Return the clause that is in force, not every version of it

    Store the contracts in a Mixedbread store and search with reranking on, adding an instruction such as preferring the latest amendment. Mixedbread's own example traces a liability cap through three amendments; at $7.50 per 1,000 queries, 50,000 searches a month cost about $375.

  • An engineer building memory for a personal assistant

    Answer from the user's current preferences when older notes contradict them

    Rerank recalled notes with v3.1 so a recent change, such as a new diet, outranks the old entry. The Starter plan's $5 credit covers about 660 reranked searches for a prototype.

  • A search lead at a 200-person company weighing a move off a self-run Elasticsearch

    Decide whether hosted search with listwise reranking justifies migrating

    Upload a sample corpus to the Scale plan at $20 a month and compare answers with and without reranking on real queries. The trade is lock-in: the reranker cannot be called from the existing cluster.

Before you commit

Setup
Create a Mixedbread store, upload files, then call stores.search with search_options.rerank set to true; v3.1 is the default model. Requests naming the retired v3 model are served by v3.1.
Integrations
Mixedbread Python and TypeScript SDKs and a REST API; Vercel billing integration. No standalone rerank endpoint for external search results.
Data and privacy
Documents are stored and indexed by Mixedbread; with bring-your-own-bucket on Enterprise, content stays in your own storage and Mixedbread says it searches it without retaining it. Certifications are not listed on the pricing page.
Cost at scale
$7.50 per 1,000 searches with reranking, so a million a month is about $7,500, plus indexing at $0.25 or $0.50 per million tokens and storage at $0.10 per million tokens a month. Enterprise volume discounts are negotiated.
Limits
Starter allows 100 queries a minute and Scale 2,400. The rerank top_k cannot exceed the first-stage count; no maximum context is published, though Mixedbread reports latency on inputs up to 128,000 tokens.

What’s good

  • Reads the whole shortlist, so superseded documents drop down
  • Follows natural-language ranking instructions
  • 0.71 NDCG@10 on ViDoRe v3 at 0.65 s, by Mixedbread's figures
  • Content-based scores make thresholds usable

What’s not

  • Only inside Mixedbread Search; no standalone rerank API
  • Weights, size and context limit are not published
  • Every benchmark is Mixedbread's own
  • Slower per query than pointwise rerankers by design

Best for

Teams building search over email, contracts or records where newer documents override older ones, and who are happy to host their corpus with Mixedbread.

mxbai-rerank-v3.1-listwise alternatives

mxbai-rerank-v3.1-listwise: common questions

Is mxbai-rerank-v3.1-listwise worth it?
A listwise reranker that reasons about which document wins, at a fraction of an LLM's latency, as long as you are willing to keep your documents in Mixedbread's search. We rate it 3.8 out of 5.
How much does mxbai-rerank-v3.1-listwise cost?
Billed as part of Mixedbread Search: semantic search $4 per 1,000 queries, plus $3.50 with reranking ($7.50 total). Indexing from $0.25 and storage $0.10 per million tokens a month. Free Starter with $5 one-time credit; Scale $20/month in credits.
What are the downsides of mxbai-rerank-v3.1-listwise?
Only inside Mixedbread Search; no standalone rerank API; weights, size and context limit are not published; every benchmark is Mixedbread's own; slower per query than pointwise rerankers by design.
Who is mxbai-rerank-v3.1-listwise best for?
Teams building search over email, contracts or records where newer documents override older ones, and who are happy to host their corpus with Mixedbread.
How much does mxbai-rerank-v3.1-listwise cost for a team or in production?
$7.50 per 1,000 searches with reranking, so a million a month is about $7,500, plus indexing at $0.25 or $0.50 per million tokens and storage at $0.10 per million tokens a month. Enterprise volume discounts are negotiated.
Is mxbai-rerank-v3.1-listwise safe for company data?
Documents are stored and indexed by Mixedbread; with bring-your-own-bucket on Enterprise, content stays in your own storage and Mixedbread says it searches it without retaining it. Certifications are not listed on the pricing page.
What are the best mxbai-rerank-v3.1-listwise alternatives?
The closest alternatives we have reviewed are mxbai-rerank v2, Cohere Rerank 5, zerank-2, Jina Reranker v3.5 and Voyage rerank-3.

Answers checked Nov 20, 2026.