---
title: "Ranking Beyond Binary Relevance: mxbai-rerank-v3-listwise"
description: Announcing mxbai-rerank-v3-listwise, our new listwise reranker codesigned with Wholembed v3. It improves results on every benchmark we ran, with state-of-the-art instruction following.
date: 2026-05-08
authors:
  - Aamir Shakir
  - Benjamin Clavié
  - Rui Huang
tags:
  - Research  
  - Product
tag: blog
url: /blog/listwise-rerank
featured: false
---

When we released [Wholembed v3](https://mixedbread.com/blog/wholembed-v3), the retrieval quality bar moved up sharply enough that most rerankers stopped helping. Pointwise rerankers, which score documents independently and sort by score, struggled to add value on top of strong first-stage results. On harder corpora, they actively hurt them.

Today we are releasing **mxbai-rerank-v3-listwise**, a listwise reranking model co-designed with Wholembed v3. It is the first reranker we have shipped that improves results on every domain, language, and benchmark we tested, lifting **all 56 Vidore v3 runs, +11% NDCG@10** on average. It also brings strong instruction-following capabilities. You can steer ranking with natural-language directives, like prioritizing recent documents, preferring internal sources, or resolving conflicts between knowledge bases.

mxbai-rerank-v3-listwise is available in preview today as part of Mixedbread Search.

## Better Reranking for Stronger Retrieval

On ViDoRe v3, a benchmark for real-world document retrieval, Wholembed v3 alone averaged 0.603 NDCG@10.[^1] Adding mxbai-rerank-v3-listwise lifted every one of the 56 runs, with an average gain of 11%.

> **Figure**: mxbai-rerank-v3-listwise lifts NDCG@10 on all 56 ViDoRe v3 runs

| Setting | NDCG@10 (avg, 56 runs) | Δ vs. retrieval-only |
| :---- | ----: | ----: |
| Wholembed v3 only | 0.603 | \- |
| \+ mxbai-rerank-v3-listwise | 0.669 | \+10.92% |

[^1]: We use ViDoRe v3 because it is a widely used benchmark for real-world document retrieval over complex corpora, with multilingual and domain-specific subsets. The benchmark covers seven domains (computer science, energy, finance, HR, industrial, pharmaceuticals, physics) across seven languages, for 56 paired runs in total.

The biggest gains came on the harder, lower-baseline subsets: industrial documents in German went up 18.8%, HR in French 16.3%. These are the kinds of corpora where small ranking errors compound into worse downstream answers.

## Comparative Ranking with Instruction Following

mxbai-rerank-v3-listwise is a listwise reranker. Unlike pointwise rerankers, which focus on the relevance of individual documents, it reads the candidate set as a whole and can resolve conflicts between candidates.

With strong instruction-following capabilities, you can tell it to prefer newer documents, prioritize internal sources over external summaries, or favor primary sources over commentary.

A booking confirmation may be superseded by a later cancellation. A product spec may override a launch note. A financial filing may be more authoritative than a same-day article.

We benchmarked mxbai-rerank-v3-listwise against the strongest pointwise rerankers available on a 900-example instruction-following evaluation covering recency-aware ranking, source-priority resolution, and multi-step composite instructions.[^2]

| Reranker | MRR | Accuracy@1 |
| ----- | ----- | ----- |
| **mxbai-rerank-v3-listwise** | **0.93** | **88.6%** |
| Voyage rerank-2.5 | 0.84 | 77.4% |
| Cohere Rerank 4 Pro | 0.77 | 68.4% |
| ZeroEntropy zerank-2 | 0.71 | 60.3% |

[^2]: The evaluation is derived from real user corpora and search patterns, then converted into controlled ranking tasks. Examples are designed to test whether the reranker can resolve conflicts inside a candidate set, not just match query-document relevance.

Much of the gap comes from recency-aware ranking, where the model has to understand that a March schedule change supersedes a January booking confirmation, or that Q1 FY27 guidance supersedes Q4 FY26 guidance.

## How Listwise Changes the Answer

To make the difference concrete, here are five cases where pointwise and listwise reranking diverge under an explicit instruction.

**Email:**
Pointwise scoring rewards the original BA confirmation: it has the highest "BA flight to London" keyword density and looks unsuperseded on its own. Listwise reads the inbox as a chain (confirmation, reschedule, cancellation, then a new booking on a different carrier) and surfaces the United flight.

**Query:** "What's the current status of my flight to London?"

**Instruction:** Resolve booking changes chronologically; the most recent valid booking wins.

| Rank | Pointwise (before) | Listwise (after) |
| --- | --- | --- |
| 1 | BA286 SFO→LHR booking confirmation (Jan 12) | UA901 SFO→LHR booking confirmation (Feb 22) |
| 2 | BA286 schedule change → 21:30 (Feb 3) | BA286 cancelled, refund processed (Feb 20) |
| 3 | UA901 SFO→LHR booking confirmation (Feb 22) | BA286 schedule change → 21:30 (Feb 3) |
| 4 | BA286 cancelled, refund processed (Feb 20) | BA286 SFO→LHR booking confirmation (Jan 12) |
| 5 | Hopper: cheap flights to London (Mar 4) | Hopper: cheap flights to London (Mar 4) |

**Finance:**
The Q3 FY26 call is the same speaker, same format, and same topic match, a strong pointwise signal, but the guidance it issues has been superseded. The Morgan Stanley note is more recent but is analyst commentary, not company guidance. Listwise routes around both and lands on the Q4 FY26 call as the current primary source.

**Query:** "What is NVDA's most recent forward revenue guidance?"

**Instruction:** Prefer primary company guidance; later filings supersede earlier; demote analyst commentary.

| Rank | Pointwise (before) | Listwise (after) |
| --- | --- | --- |
| 1 | Q3 FY26 call — Q4 FY26 guidance ($37.5B) (Nov 20 '25) | Q4 FY26 call — Q1 FY27 guidance ($43B) (Feb 26 '26) |
| 2 | Morgan Stanley NVDA note (Q1 model) (Mar 15 '26) | CNBC: "Nvidia guides Q1 above expectations" (Feb 26 '26) |
| 3 | Q4 FY26 call — Q1 FY27 guidance ($43B) (Feb 26 '26) | Morgan Stanley NVDA note (Q1 model) (Mar 15 '26) |
| 4 | CNBC: "Nvidia guides Q1 above expectations" (Feb 26 '26) | Q3 FY26 call — Q4 FY26 guidance ($37.5B) (Nov 20 '25) |
| 5 | NVDA 8-K — data center supply chain (Apr 2 '26) | NVDA 8-K — data center supply chain (Apr 2 '26) |

**Legal:**
The Code of Conduct is keyword-dense for "indemnification" but belongs to a different agreement; Amendment 4 is the most recent document but doesn't touch §12. Listwise traces the §12 chain (base → scope rewrite → cap reset) to identify Amendment 3 as the operative cap, sitting on top of Amendment 2's scope.

**Query:** "Find the operative indemnification provision."

**Instruction:** Trace Section 12 amendments; ignore amendments and docs that don't touch this section.

| Rank | Pointwise (before) | Listwise (after) |
| --- | --- | --- |
| 1 | Code of Conduct (boilerplate indemn.) (2024-02) | Amendment 3 — §12 cap → $5M (2025-07) |
| 2 | MSA §12 — base indemnification (2022-06) | Amendment 2 — §12 scope rewrite (2023-11) |
| 3 | Amendment 2 — §12 scope rewrite (2023-11) | MSA §12 — base indemnification (2022-06) |
| 4 | Amendment 3 — §12 cap → $5M (2025-07) | Amendment 4 — §8 only (term) (2026-01) |
| 5 | Amendment 4 — §8 only (term) (2026-01) | Code of Conduct (boilerplate indemn.) (2024-02) |

**Memory:**
Pointwise ranks dietary snippets by topic match, mixing current and superseded statements together. Listwise reads the chain (vegetarian → pescatarian → plant-based → pescatarian) and treats the peanut allergy as an orthogonal additive constraint, surfacing both the current pescatarian statement and the additive allergy statement at the top.

**Query:** "What are the user's current dietary restrictions?"

**Instruction:** Synthesize current state; later statements supersede earlier ones; allergies are additive.

| Rank | Pointwise (before) | Listwise (after) |
| --- | --- | --- |
| 1 | "Vegetarian for about five years." (Aug 2025) | "Plant-based didn't stick — back to pescatarian." (Apr 2026) |
| 2 | "Going plant-based for a 90-day reset." (Feb 2026) | "Diagnosed mild peanut allergy." (Jan 2026) |
| 3 | "Started eating fish again." (Oct 2025) | "Going plant-based for a 90-day reset." (Feb 2026) |
| 4 | "Plant-based didn't stick — back to pescatarian." (Apr 2026) | "Started eating fish again." (Oct 2025) |
| 5 | "Diagnosed mild peanut allergy." (Jan 2026) | "Vegetarian for about five years." (Aug 2025) |

**Models:**
The benchmark explainer is the most "MRR + reranker" keyword-dense candidate, so the pointwise model promotes it, even though it is not a measured reranker row. Listwise applies the instruction to compare results from the same evaluation and ranks the model with the highest reported MRR first.

**Query:** "Which reranker has the highest MRR?"

**Instruction:** Compare only rows from the same instruction-following evaluation; sort by MRR descending.

| Rank | Pointwise (before) | Listwise (after) |
| --- | --- | --- |
| 1 | Blog: "how MRR evaluates rerankers" (n/a) | mxbai-rerank-v3-listwise benchmark row (0.93 MRR) |
| 2 | Voyage rerank-2.5 benchmark row (0.84 MRR) | Voyage rerank-2.5 benchmark row (0.84 MRR) |
| 3 | Cohere Rerank 4 Pro benchmark row (0.77 MRR) | Cohere Rerank 4 Pro benchmark row (0.77 MRR) |
| 4 | mxbai-rerank-v3-listwise benchmark row (0.93 MRR) | ZeroEntropy zerank-2 benchmark row (0.71 MRR) |
| 5 | ZeroEntropy zerank-2 benchmark row (0.71 MRR) | Blog: "how MRR evaluates rerankers" (n/a) |

## One API

mxbai-rerank-v3-listwise is available in preview today through Mixedbread Search:

**Python:**
```python
from mixedbread import Mixedbread

    client = Mixedbread()

    results = client.stores.search(
        store_identifiers=["my-store"],
        query="when is my flight to London? The most recent valid booking wins",
        search_options={
          "rerank": {
              "model": "mixedbread-ai/mxbai-rerank-v3-listwise",
          }
        },
    )
```

**TypeScript:**
```typescript
import { Mixedbread } from "@mixedbread/sdk";

    const client = new Mixedbread();

    const results = await client.stores.search({
        store_identifiers: ["my-store"],
        query: "when is my flight to London? The most recent valid booking wins",
        search_options: {
          rerank: {
              model: "mixedbread-ai/mxbai-rerank-v3-listwise",
          }
        },
    });
```

New users get $5 free credits to [try it](https://www.platform.mixedbread.com/).

If this is a problem you want to work on, we are [hiring](https://mixedbread.com/careers).