🏆 Headline results — specialized beats generalist at reranking

On a held-out test of 3,157 queries over 1,534 scientific papers (9-way reranking with hard negatives), this 149M-parameter model picks the correct document first 93% of the time — and lands it in the top 3 a remarkable 98.2% of the time — outscoring OpenAI's general embeddings, keyword search, and hybrid retrieval, at 41 ms/query on hardware you can self-host.

Method	Precision@1	Precision@3	MRR
Dense embeddings — OpenAI text-embedding-3-small	0.681	—	0.745
Hybrid retrieval (RRF, k=60)	0.728	—	0.788
BM25 (keyword search)	0.856	—	0.905
44b-reranker-minilm-l6 (22.7M, fast tier)	0.865	0.954	0.914
44b-reranker-gte-modernbert (149.6M — this model)	0.929	0.982	0.956

What that means, all measured on the same test set:

• 93% vs 68% over OpenAI's general embeddings on top-1 accuracy (+25 points); 95.6% vs 74.5% on MRR.
• 93% vs 86% over BM25 keyword search; 93% vs 73% over hybrid retrieval.
• Right document in the top 3: 98.2% — up from 92.6% for the general-purpose base.
• 57% fewer ranking errors than the base it is fine-tuned from (16.7% → 7.1%). Mean rank 1.148 of 9.

Scope: reranking numbers — selecting the right document from a candidate set — not open-corpus search across millions. P@3 for the first-stage retrieval arms was not reported. A companion 22.7M model (44b-reranker-minilm-l6) trades ~6 points of accuracy for ~6× the speed.

---

44b-reranker-gte-modernbert

A cross-encoder reranker fine-tuned on peer review → paper pairs mined from the 44B corpus.

The supervision is unusual and it is the point: the query is a peer reviewer's own summary of a paper, written by a domain expert who had never heard of this model, and the positive is the paper that review is attached to. The relevance label is not a heuristic, not a click, and not LLM-generated.

Fine-tuning on it cut the base model's error rate by 57% on a held-out, paper-disjoint evaluation.

Results

Evaluated on 3,157 held-out queries over 1,534 papers, none of which appear anywhere in training — not as positives and not as negatives. Each query is scored against 9 candidates: 1 positive and 8 mined hard negatives. Random chance is P@1 = 0.1111.

model	P@1	P@3	MRR	mean rank	ms/query
this model	0.9290	0.9816	0.9563	1.148	41.1
gte-reranker-modernbert-base (base)	0.8334	0.9262	0.8888	1.464	53.1
chance	0.1111	0.3333	—	5.000	—

Error rate 16.66% → 7.10%, a 57.4% relative reduction. At n = 3,157 the gap is far beyond noise (z ≈ 11.9 on an unpaired standard error, which is conservative — these are paired measurements on identical queries).

For reference, the first-stage retrieval arms scored on the same 3,157 queries:

first stage	P@1	MRR
BM25 (lexical)	0.8562	0.9046
Hybrid RRF (k=60)	0.7282	0.7881
Dense (text-embedding-3-small)	0.6807	0.7450

A sibling model, NYSgpt/44b-reranker-minilm-l6, was trained on the identical data and reaches P@1 0.8654 at 7.4 ms/query — 1/7th the parameters and ~6× the speed. If latency matters more than the last three points of accuracy, use that one.

Read the eval shape before you read the numbers

This is a 9-way ranking task, not open-corpus retrieval. A 0.9290 here means "picks the right paper out of nine confusable ones," not "finds the right paper in three million." The negatives are hard by construction — nearest neighbours in embedding space plus lexical near-matches — but the pool is small and the task is closed.

What it was trained on

training groups	24,177
negatives per group	7
distinct papers in train	12,058
held out (val + test)	3,099 papers
query	an official review's summary, or a meta-review's metareview
positive	the paper the review is attached to
negatives	mined per query — semantic (HNSW over text-embedding-3-small, ef_search=160) and lexical (Postgres FTS over abstracts)
source venues	ICLR 2024, ICLR 2025, NeurIPS 2024

Splits are disjoint by paper, not by pair. Papers carry ~4.5 reviews each, so splitting by pair would place the same paper in train and test and inflate every number downstream. Proven at build time rather than asserted: |train ∩ val| = 0, |train ∩ test| = 0, |val ∩ test| = 0.

Held-out papers are excluded from the negative pool as well as the positive one. Label leakage was measured after the fact, not assumed: 0 of 1,534 test positives appear as training positives; 23 (1.5%) appear as a negative for some other query, which biases mildly against this model rather than for it.

Training

base	Alibaba-NLP/gte-reranker-modernbert-base (150M params)
objective	grouped cross-entropy — one positive against 7 hard negatives per group
epochs	2
max sequence length	512
query / document truncation	600 / 900 characters
learning rate	2e-5
hardware	1× NVIDIA A10G (g5.xlarge)
wall clock	83 minutes
training loss	0.4734 → 0.0883

Limitations

Stated plainly, because most of these are the kind of thing a reader would otherwise have to discover.

• Query register is long-form prose. Training and evaluation queries average 700 characters (174 tokens) of fluent reviewer writing. This model has never seen a short keyword query. Do not assume the gains transfer to a two-word search box — that is untested.
• One discipline, one document type. Every review comes from top-tier machine learning conferences (ICLR, NeurIPS; 2024–2025). No other field, venue tier, or document class is represented.
• Context is 512, not 8192. The base supports long context; this fine-tune only ever saw 512 tokens, so the extra capacity is untrained.
• Negatives skew semantic. 92% of mined negatives came from the embedding arm and 8% from the lexical arm — the lexical miner conjoined a review's rarest terms and returned nothing on most queries, so the semantic arm backfilled. A balanced mine is the obvious next version.
• Trained on 7 negatives, evaluated against 8. Evaluation is marginally harder than training was.
• The evaluation shares a pipeline with the training data. Same source, same construction, disjoint only by paper. It measures the reranker's job honestly; it is not an independent measure of end-to-end answer quality in a live system.
• CPU latency is unmeasured. Every timing here is A10G at batch 128. No CPU serving number exists yet.
• False-negative screening may under-catch. Candidate negatives were rejected for sharing the positive's Semantic Scholar id or a near-identical title; the rate was 0.118%, and the S2-duplicate check fired zero times across 28,798 candidates. That is either a clean corpus or a check that is not working, and we have not yet established which.
• The supervision is not exhausted. This model trained on 30,509 of 77,147 available pairs. It is a first cut, not a ceiling.

Intended use

Second-stage reranking of scientific-document retrieval — reordering a first-stage candidate list for queries written in the register of a paragraph of technical prose.

Out of scope

Short keyword queries · non-ML disciplines · any setting requiring a measured CPU latency guarantee · any claim of quality on open-corpus retrieval, which this evaluation does not test.

Provenance

Reviews from OpenReview; papers from arXiv, OpenAlex and Semantic Scholar, unified in the 44B corpus. Trained 2026-07-31.