rewirebio.iobenchmarks
Evidence and investigation

Contested results

Where one study disagrees with what other studies report for the same models, and what we are doing to resolve each case.

These are automatic leads, not findings of error. Each release, a check compares how every study ranks the models it shares with other studies. It flags a study that reverses the order the others agree on, or whose gap to a shared reference model lies well outside theirs. Different tasks, populations or metrics often explain a flag.

A flag never hides a result. Flagged results stay in every table, page and download, marked Contested. A result is withheld only when a reviewer marks it disputed.

How the check works, and how each follow-up is recorded: discrepancies between studies.

In this release

  • 192open
  • 6settled
  • 7resolved
  • 3medium severity
  • 202low severity

Release 2026-11-21-adc15377fab2 · 205 findings in all · rules version 5

Findings

Status: All · Open · Settled · Resolved

Severity: All · Medium · Low

State: All · Not yet reviewed · Under investigation · Explained · Confirmed error · Not a real disagreement

Showing findings that involve results in Select regulatory variants and genes for functional follow-up, from the study that disagrees or from the agreeing group. All use cases

12 findings match, most severe first.

Caduceus and GENA-LM: GenBench: A Benchmarking Suite for Systematic Evaluation of Genomic Foundation Models reverses the order

  • Low severity
  • Not yet reviewed
  • Ordering reversal
  • unknown

GenBench: A Benchmarking Suite for Systematic Evaluation of Genomic Foundation Models ranks GENA-LM above Caduceus in 41 of 48 comparisons; 2 of 2 other studies rank Caduceus above GENA-LM.

3 papers rank Caduceus above GENA-LM in 23 of their 55 comparisons and GENA-LM higher in 31; 1 is a tie. Of those papers, GENEB leaderboard ranks GENA-LM higher on its own (30 of 45). GenBench ranks GENA-LM above Caduceus in 41 of its 51 comparisons and Caduceus higher in 7; 3 are ties.

Comparisons in which each paper scores Caduceus or GENA-LM higher3 other papers: Caduceus higher in 23, GENA-LM higher in 31, 1 tie. GenBench (disagrees): Caduceus higher in 7, GENA-LM higher in 41, 3 ties.3 other papers23 Caduceus higher231 tie31 GENA-LM higher31Disagrees: GenBenchGenBench: A Benchmarking Suite for Systematic Evaluation of Genomic Foundation Models7 Caduceus higher73 tie41 GENA-LM higher41
Caduceus higherTieGENA-LM higher
Compared
Caduceus, GENA-LM

Not reviewed yet.

DNABERT-2 and Nucleotide Transformer: GenomeOcean: An Efficient Genome Foundation Model Trained on Large-Scale Metagenomic Assemblies reverses the order

  • Low severity
  • Not yet reviewed
  • Ordering reversal
  • unknown

GenomeOcean: An Efficient Genome Foundation Model Trained on Large-Scale Metagenomic Assemblies ranks DNABERT-2 above Nucleotide Transformer in 3 of 3 comparisons; 7 of 8 other studies rank Nucleotide Transformer above DNABERT-2.

10 papers rank Nucleotide Transformer above DNABERT-2 in 104 of their 171 comparisons and DNABERT-2 higher in 52; 15 are ties. Of those papers, DNABERT-2, BEND rank DNABERT-2 higher on their own (24 of 28, 5 of 7). GenomeOcean 2025 ranks DNABERT-2 above Nucleotide Transformer in all 3 of its comparisons.

Comparisons in which each paper scores Nucleotide Transformer or DNABERT-2 higher10 other papers: Nucleotide Transformer higher in 104, DNABERT-2 higher in 52, 15 ties. GenomeOcean 2025 (disagrees): Nucleotide Transformer higher in 0, DNABERT-2 higher in 3, 0 ties.10 other papers104 Nucleotide Transformer higher10415 tie52 DNABERT-2 higher52Disagrees: GenomeOcean 2025GenomeOcean: An Efficient Genome Foundation Model Trained on Large-Scale Metagenomic Assemblies3 DNABERT-2 higher3
Nucleotide Transformer higherTieDNABERT-2 higher
Compared
DNABERT-2, Nucleotide Transformer

Not reviewed yet.

DNABERT-2 and Nucleotide Transformer: DNABERT-2: Efficient Foundation Model and Benchmark for Multi-Species Genomes reverses the order

  • Low severity
  • Not yet reviewed
  • Ordering reversal
  • unknown

DNABERT-2: Efficient Foundation Model and Benchmark for Multi-Species Genomes ranks DNABERT-2 above Nucleotide Transformer in 24 of 26 comparisons; 7 of 8 other studies rank Nucleotide Transformer above DNABERT-2.

10 papers rank Nucleotide Transformer above DNABERT-2 in 102 of their 146 comparisons and DNABERT-2 higher in 31; 13 are ties. Of those papers, BEND, GenomeOcean 2025 rank DNABERT-2 higher on their own (5 of 7, 3 of 3). DNABERT-2 ranks DNABERT-2 above Nucleotide Transformer in 24 of its 28 comparisons and Nucleotide Transformer higher in 2; 2 are ties.

Comparisons in which each paper scores Nucleotide Transformer or DNABERT-2 higher10 other papers: Nucleotide Transformer higher in 102, DNABERT-2 higher in 31, 13 ties. DNABERT-2 (disagrees): Nucleotide Transformer higher in 2, DNABERT-2 higher in 24, 2 ties.10 other papers102 Nucleotide Transformer higher10213 tie31 DNABERT-2 higher31Disagrees: DNABERT-2DNABERT-2: Efficient Foundation Model and Benchmark for Multi-Species Genomes2 Nucleotide Transformer higher2 tie24 DNABERT-2 higher24
Nucleotide Transformer higherTieDNABERT-2 higher
Compared
DNABERT-2, Nucleotide Transformer

Not reviewed yet.

Caduceus and HyenaDNA: A Phylogenetic Approach to Genomic Language Modeling reverses the order

  • Low severity
  • Not yet reviewed
  • Ordering reversal
  • unknown

A Phylogenetic Approach to Genomic Language Modeling ranks Caduceus above HyenaDNA in 9 of 12 comparisons; 2 of 2 other studies rank HyenaDNA above Caduceus.

5 papers rank HyenaDNA above Caduceus in 63 of their 133 comparisons and Caduceus higher in 49; 21 are ties. Of those papers, DNALONGBENCH ranks Caduceus higher on its own (16 of 27). A Phylogenetic Approach to Genomic… ranks Caduceus above HyenaDNA in 9 of its 12 comparisons and HyenaDNA higher in 3.

Comparisons in which each paper scores HyenaDNA or Caduceus higher5 other papers: HyenaDNA higher in 63, Caduceus higher in 49, 21 ties. A Phylogenetic Approach to Genomic… (disagrees): HyenaDNA higher in 3, Caduceus higher in 9, 0 ties.5 other papers63 HyenaDNA higher6321 tie2149 Caduceus higher49Disagrees: A Phylogenetic Approach to Genomic…A Phylogenetic Approach to Genomic Language Modeling3 HyenaDNA higher39 Caduceus higher9
HyenaDNA higherTieCaduceus higher
Compared
Caduceus, HyenaDNA

Not reviewed yet.

Evo 2 and phyloP: TraitGym official metrics: AUPRC by chromosome, weighted average (Hugging Face dataset songlab/TraitGym) reverses the order

  • Low severity
  • Not yet reviewed
  • Ordering reversal
  • different-task

TraitGym official metrics: AUPRC by chromosome, weighted average (Hugging Face dataset songlab/TraitGym) ranks phyloP above Evo 2 in 2 of 2 comparisons; 2 of 2 other studies rank Evo 2 above phyloP.

3 papers rank Evo 2 above phyloP in 83 of their 107 comparisons and phyloP higher in 16; 8 are ties. TraitGym official metrics ranks phyloP above Evo 2 in all 2 of its comparisons.

Comparisons in which each paper scores Evo 2 or phyloP higher3 other papers: Evo 2 higher in 83, phyloP higher in 16, 8 ties. TraitGym official metrics (disagrees): Evo 2 higher in 0, phyloP higher in 2, 0 ties.3 other papers83 Evo 2 higher838 tie16 phyloP higher16Disagrees: TraitGym official metricsTraitGym official metrics: AUPRC by chromosome, weighted average (Hugging Face dataset songlab/TraitGym)2 phyloP higher2
Evo 2 higherTiephyloP higher
Compared
Evo 2, phyloP

Not reviewed yet.

DNABERT-2 and GENA-LM: GUANinE v1.1 reveals complementarity of supervised and genomic language models reverses the order

  • Low severity
  • Not yet reviewed
  • Ordering reversal
  • unknown

GUANinE v1.1 reveals complementarity of supervised and genomic language models ranks GENA-LM above DNABERT-2 in 6 of 6 comparisons; 2 of 2 other studies rank DNABERT-2 above GENA-LM.

4 papers rank DNABERT-2 above GENA-LM in 62 of their 112 comparisons and GENA-LM higher in 26; 24 are ties. GUANinE v1.1 reveals complementarity… ranks GENA-LM above DNABERT-2 in all 6 of its comparisons.

Comparisons in which each paper scores DNABERT-2 or GENA-LM higher4 other papers: DNABERT-2 higher in 62, GENA-LM higher in 26, 24 ties. GUANinE v1.1 reveals complementarity… (disagrees): DNABERT-2 higher in 0, GENA-LM higher in 6, 0 ties.4 other papers62 DNABERT-2 higher6224 tie2426 GENA-LM higher26Disagrees: GUANinE v1.1 reveals complementarity…GUANinE v1.1 reveals complementarity of supervised and genomic language models6 GENA-LM higher6
DNABERT-2 higherTieGENA-LM higher
Compared
DNABERT-2, GENA-LM

Not reviewed yet.

Nucleotide Transformer and HyenaDNA: TraitGym official metrics: AUPRC by chromosome, weighted average (Hugging Face dataset songlab/TraitGym) reverses the order

  • Low severity
  • Not yet reviewed
  • Ordering reversal
  • unknown

TraitGym official metrics: AUPRC by chromosome, weighted average (Hugging Face dataset songlab/TraitGym) ranks HyenaDNA above Nucleotide Transformer in 2 of 2 comparisons; 4 of 4 other studies rank Nucleotide Transformer above HyenaDNA.

5 papers rank Nucleotide Transformer above HyenaDNA in 77 of their 82 comparisons and HyenaDNA higher in 5. TraitGym official metrics ranks HyenaDNA above Nucleotide Transformer in all 2 of its comparisons.

Comparisons in which each paper scores Nucleotide Transformer or HyenaDNA higher5 other papers: Nucleotide Transformer higher in 77, HyenaDNA higher in 5, 0 ties. TraitGym official metrics (disagrees): Nucleotide Transformer higher in 0, HyenaDNA higher in 2, 0 ties.5 other papers77 Nucleotide Transformer higher775 HyenaDNA higherDisagrees: TraitGym official metricsTraitGym official metrics: AUPRC by chromosome, weighted average (Hugging Face dataset songlab/TraitGym)2 HyenaDNA higher2
Nucleotide Transformer higherTieHyenaDNA higher
Compared
Nucleotide Transformer, HyenaDNA

Not reviewed yet.

ChromBPNet and Caduceus: GUANinE v1.1 reveals complementarity of supervised and genomic language models and Comparative Analysis of Deep Learning Models for Predicting Causative Regulatory Variants disagree

  • Low severity
  • Not yet reviewed
  • Two-study conflict
  • unknown

GUANinE v1.1 reveals complementarity of supervised and genomic language models ranks Caduceus above ChromBPNet in 6 of 6 comparisons; Comparative Analysis of Deep Learning Models for Predicting Causative Regulatory Variants ranks ChromBPNet above Caduceus in 4 of 4 comparisons. No other study gives a verdict, so neither is the outlier.

2 papers rank ChromBPNet above Caduceus in 4 of their 10 comparisons and Caduceus higher in 6. Of those papers, GUANinE v1.1 reveals complementarity… ranks Caduceus higher on its own (6 of 6).

Comparisons in which each paper scores ChromBPNet or Caduceus higher2 other papers: ChromBPNet higher in 4, Caduceus higher in 6, 0 ties.2 other papers4 ChromBPNet higher46 Caduceus higher6
ChromBPNet higherTieCaduceus higher
Compared
ChromBPNet, Caduceus

Not reviewed yet.

Evo 2 and Caduceus: GUANinE v1.1 reveals complementarity of supervised and genomic language models and GFMBench-API: A Standardized Interface for Benchmarking Genomic Foundation Models disagree

  • Low severity
  • Not yet reviewed
  • Two-study conflict
  • different-task

GUANinE v1.1 reveals complementarity of supervised and genomic language models ranks Caduceus above Evo 2 in 4 of 4 comparisons; GFMBench-API: A Standardized Interface for Benchmarking Genomic Foundation Models ranks Evo 2 above Caduceus in 35 of 46 comparisons. No other study gives a verdict, so neither is the outlier.

3 papers rank Evo 2 above Caduceus in 36 of their 58 comparisons and Caduceus higher in 15; 7 are ties. Of those papers, GUANinE v1.1 reveals complementarity… ranks Caduceus higher on its own (4 of 6).

Comparisons in which each paper scores Evo 2 or Caduceus higher3 other papers: Evo 2 higher in 36, Caduceus higher in 15, 7 ties.3 other papers36 Evo 2 higher367 tie715 Caduceus higher15
Evo 2 higherTieCaduceus higher
Compared
Evo 2, Caduceus

Not reviewed yet.

Borzoi and Sei: GUANinE v1.1 reveals complementarity of supervised and genomic language models and Comparative Analysis of Deep Learning Models for Predicting Causative Regulatory Variants disagree

  • Low severity
  • Not yet reviewed
  • Two-study conflict
  • unknown

GUANinE v1.1 reveals complementarity of supervised and genomic language models ranks Borzoi above Sei in 6 of 6 comparisons; Comparative Analysis of Deep Learning Models for Predicting Causative Regulatory Variants ranks Sei above Borzoi in 4 of 4 comparisons. No other study gives a verdict, so neither is the outlier.

2 papers rank Borzoi above Sei in 6 of their 11 comparisons and Sei higher in 4; 1 is a tie. Of those papers, Comparative Analysis of Deep Learning… ranks Sei higher on its own (4 of 4).

Comparisons in which each paper scores Borzoi or Sei higher2 other papers: Borzoi higher in 6, Sei higher in 4, 1 tie.2 other papers6 Borzoi higher61 tie14 Sei higher4
Borzoi higherTieSei higher
Compared
Borzoi, Sei

Not reviewed yet.

ChromBPNet and GENA-LM: GUANinE v1.1 reveals complementarity of supervised and genomic language models and Comparative Analysis of Deep Learning Models for Predicting Causative Regulatory Variants disagree

  • Low severity
  • Not yet reviewed
  • Two-study conflict
  • unknown

GUANinE v1.1 reveals complementarity of supervised and genomic language models ranks GENA-LM above ChromBPNet in 3 of 4 comparisons; Comparative Analysis of Deep Learning Models for Predicting Causative Regulatory Variants ranks ChromBPNet above GENA-LM in 4 of 4 comparisons. No other study gives a verdict, so neither is the outlier.

2 papers rank ChromBPNet above GENA-LM in 5 of their 10 comparisons and GENA-LM higher in 3; 2 are ties. Of those papers, GUANinE v1.1 reveals complementarity… ranks GENA-LM higher on its own (3 of 6).

Comparisons in which each paper scores ChromBPNet or GENA-LM higher2 other papers: ChromBPNet higher in 5, GENA-LM higher in 3, 2 ties.2 other papers5 ChromBPNet higher52 tie23 GENA-LM higher3
ChromBPNet higherTieGENA-LM higher
Compared
ChromBPNet, GENA-LM

Not reviewed yet.

ChromBPNet and Borzoi: GUANinE v1.1 reveals complementarity of supervised and genomic language models and Comparative Analysis of Deep Learning Models for Predicting Causative Regulatory Variants disagree

  • Low severity
  • Not yet reviewed
  • Two-study conflict
  • unknown

GUANinE v1.1 reveals complementarity of supervised and genomic language models ranks Borzoi above ChromBPNet in 5 of 5 comparisons; Comparative Analysis of Deep Learning Models for Predicting Causative Regulatory Variants ranks ChromBPNet above Borzoi in 4 of 4 comparisons. No other study gives a verdict, so neither is the outlier.

3 papers rank ChromBPNet above Borzoi in 7 of their 21 comparisons and Borzoi higher in 12; 2 are ties. Of those papers, AlphaGenome Nature 2026, GUANinE v1.1 reveals complementarity… rank Borzoi higher on their own (7 of 10, 5 of 7).

Comparisons in which each paper scores ChromBPNet or Borzoi higher3 other papers: ChromBPNet higher in 7, Borzoi higher in 12, 2 ties.3 other papers7 ChromBPNet higher72 tie212 Borzoi higher12
ChromBPNet higherTieBorzoi higher
Compared
ChromBPNet, Borzoi

Not reviewed yet.