{ "schema_version": 1, "project": "GenomicTouchstone", "updated": "2026-10-19", "description": "Papers and released model resources for the 63 models and methods in the revised study. Evaluation modes apply to their respective task subsets.", "counts": { "Genomic LMs": 32, "RNA LMs": 6, "Protein LMs": 11, "Task-specific": 10, "Baselines": 4 }, "models": [ { "id": "dnabert-3", "name": "DNABERT (3-mer)", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "dnabert-3", "aliases": [ "DNABERT-3mer" ], "note": "", "paper": { "url": "https://doi.org/10.1093/bioinformatics/btab083", "label": "Paper", "title": "DNABERT: pre-trained Bidirectional Encoder Representations from Transformers model for DNA-language in genome" }, "download": { "url": "https://huggingface.co/zhihan1996/DNA_bert_3", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/zhihan1996/DNA_bert_3", "publication": { "first_public_date": "2020-09-19", "first_public_url": "https://doi.org/10.1101/2020.09.17.301879", "first_public_platform": "bioRxiv", "venue": "Bioinformatics", "venue_year": "2021", "venue_url": "https://doi.org/10.1093/bioinformatics/btab083", "note": "" }, "specifications": { "architecture": "Attention", "parameters": "86.1M", "input_length": "514", "tokenizer": "Overlapping 3-mer", "pretraining_objective": "MLM", "pretraining_data": "Human genome (GRCh38/hg38)", "input_length_unit": "nt" } }, { "id": "dnabert-4", "name": "DNABERT (4-mer)", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "dnabert-4", "aliases": [ "DNABERT-4mer" ], "note": "", "paper": { "url": "https://doi.org/10.1093/bioinformatics/btab083", "label": "Paper", "title": "DNABERT: pre-trained Bidirectional Encoder Representations from Transformers model for DNA-language in genome" }, "download": { "url": "https://huggingface.co/zhihan1996/DNA_bert_4", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/zhihan1996/DNA_bert_4", "publication": { "first_public_date": "2020-09-19", "first_public_url": "https://doi.org/10.1101/2020.09.17.301879", "first_public_platform": "bioRxiv", "venue": "Bioinformatics", "venue_year": "2021", "venue_url": "https://doi.org/10.1093/bioinformatics/btab083", "note": "" }, "specifications": { "architecture": "Attention", "parameters": "86.1M", "input_length": "515", "tokenizer": "Overlapping 4-mer", "pretraining_objective": "MLM", "pretraining_data": "Human genome (GRCh38/hg38)", "input_length_unit": "nt" } }, { "id": "dnabert-5", "name": "DNABERT (5-mer)", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "dnabert-5", "aliases": [ "DNABERT-5mer" ], "note": "", "paper": { "url": "https://doi.org/10.1093/bioinformatics/btab083", "label": "Paper", "title": "DNABERT: pre-trained Bidirectional Encoder Representations from Transformers model for DNA-language in genome" }, "download": { "url": "https://huggingface.co/zhihan1996/DNA_bert_5", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/zhihan1996/DNA_bert_5", "publication": { "first_public_date": "2020-09-19", "first_public_url": "https://doi.org/10.1101/2020.09.17.301879", "first_public_platform": "bioRxiv", "venue": "Bioinformatics", "venue_year": "2021", "venue_url": "https://doi.org/10.1093/bioinformatics/btab083", "note": "" }, "specifications": { "architecture": "Attention", "parameters": "86.1M", "input_length": "516", "tokenizer": "Overlapping 5-mer", "pretraining_objective": "MLM", "pretraining_data": "Human genome (GRCh38/hg38)", "input_length_unit": "nt" } }, { "id": "dnabert-6", "name": "DNABERT (6-mer)", "category": "Genomic LMs", "evaluation": "FT + LP", "leaderboard_id": "dnabert-6", "aliases": [ "DNABERT-6mer" ], "note": "", "paper": { "url": "https://doi.org/10.1093/bioinformatics/btab083", "label": "Paper", "title": "DNABERT: pre-trained Bidirectional Encoder Representations from Transformers model for DNA-language in genome" }, "download": { "url": "https://huggingface.co/zhihan1996/DNA_bert_6", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/zhihan1996/DNA_bert_6", "publication": { "first_public_date": "2020-09-19", "first_public_url": "https://doi.org/10.1101/2020.09.17.301879", "first_public_platform": "bioRxiv", "venue": "Bioinformatics", "venue_year": "2021", "venue_url": "https://doi.org/10.1093/bioinformatics/btab083", "note": "" }, "specifications": { "architecture": "Attention", "parameters": "86.1M", "input_length": "517", "tokenizer": "Overlapping 6-mer", "pretraining_objective": "MLM", "pretraining_data": "Human genome (GRCh38/hg38)", "input_length_unit": "nt" } }, { "id": "dnabert2", "name": "DNABERT-2", "category": "Genomic LMs", "evaluation": "FT + LP", "leaderboard_id": "dnabert2", "aliases": [ "DNABERT2" ], "note": "", "paper": { "url": "https://arxiv.org/abs/2306.15006", "label": "Paper", "title": "DNABERT-2: EFFICIENT FOUNDATION MODEL AND BENCHMARK FOR MULTI-SPECIES GENOMES" }, "download": { "url": "https://huggingface.co/zhihan1996/DNABERT-2-117M", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/zhihan1996/DNABERT-2-117M", "publication": { "first_public_date": "2023-06-26", "first_public_url": "https://arxiv.org/abs/2306.15006v1", "first_public_platform": "arXiv", "venue": "ICLR", "venue_year": "2024", "venue_url": "https://arxiv.org/abs/2306.15006", "note": "" }, "specifications": { "architecture": "Attention", "parameters": "117M", "input_length": "Variable", "tokenizer": "BPE", "pretraining_objective": "MLM", "pretraining_data": "Human genome + Multiple species genome", "input_length_unit": "nt" } }, { "id": "gena-lm-bert-base", "name": "GENA-BERT-Base", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "gena-lm-bert-base", "aliases": [ "Gena-LM-Base" ], "note": "", "paper": { "url": "https://doi.org/10.1093/nar/gkae1310", "label": "Paper", "title": "GENA-LM: a family of open-source foundational DNA language models for long sequences" }, "download": { "url": "https://huggingface.co/AIRI-Institute/gena-lm-bert-base", "label": "Weights", "kind": "weights" }, "source_url": "https://github.com/AIRI-Institute/GENA_LM", "publication": { "first_public_date": "2023-06-13", "first_public_url": "https://doi.org/10.1101/2023.06.12.544594", "first_public_platform": "bioRxiv", "venue": "Nucleic Acids Research", "venue_year": "2025", "venue_url": "https://doi.org/10.1093/nar/gkae1310", "note": "" }, "specifications": { "architecture": "Attention", "parameters": "110M", "input_length": "≈4,500", "tokenizer": "BPE", "pretraining_objective": "MLM", "pretraining_data": "T2T split v1", "input_length_unit": "nt" } }, { "id": "gena-lm-bert-base-lastln-t2t", "name": "GENA-BERT-Base-LastLN-T2T", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "gena-lm-bert-base-lastln-t2t", "aliases": [ "Gena-LM-Base-Lastln-T2T" ], "note": "", "paper": { "url": "https://doi.org/10.1093/nar/gkae1310", "label": "Paper", "title": "GENA-LM: a family of open-source foundational DNA language models for long sequences" }, "download": { "url": "https://huggingface.co/AIRI-Institute/gena-lm-bert-base-lastln-t2t", "label": "Weights", "kind": "weights" }, "source_url": "https://github.com/AIRI-Institute/GENA_LM", "publication": { "first_public_date": "2023-06-13", "first_public_url": "https://doi.org/10.1101/2023.06.12.544594", "first_public_platform": "bioRxiv", "venue": "Nucleic Acids Research", "venue_year": "2025", "venue_url": "https://doi.org/10.1093/nar/gkae1310", "note": "" }, "specifications": { "architecture": "Attention", "parameters": "110M", "input_length": "≈4,500", "tokenizer": "BPE", "pretraining_objective": "MLM", "pretraining_data": "T2T + 1000G SNPs", "input_length_unit": "nt" } }, { "id": "gena-lm-bert-base-t2t", "name": "GENA-BERT-Base-T2T", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "gena-lm-bert-base-t2t", "aliases": [ "Gena-LM-Base-T2T" ], "note": "", "paper": { "url": "https://doi.org/10.1093/nar/gkae1310", "label": "Paper", "title": "GENA-LM: a family of open-source foundational DNA language models for long sequences" }, "download": { "url": "https://huggingface.co/AIRI-Institute/gena-lm-bert-base-t2t", "label": "Weights", "kind": "weights" }, "source_url": "https://github.com/AIRI-Institute/GENA_LM", "publication": { "first_public_date": "2023-06-13", "first_public_url": "https://doi.org/10.1101/2023.06.12.544594", "first_public_platform": "bioRxiv", "venue": "Nucleic Acids Research", "venue_year": "2025", "venue_url": "https://doi.org/10.1093/nar/gkae1310", "note": "" }, "specifications": { "architecture": "Attention", "parameters": "110M", "input_length": "≈4,500", "tokenizer": "BPE", "pretraining_objective": "MLM", "pretraining_data": "T2T + 1000G SNPs", "input_length_unit": "nt" } }, { "id": "gena-lm-bert-base-t2t-multi", "name": "GENA-BERT-Base-T2T-Multi", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "gena-lm-bert-base-t2t-multi", "aliases": [ "Gena-LM-Base-T2T-Multi" ], "note": "", "paper": { "url": "https://doi.org/10.1093/nar/gkae1310", "label": "Paper", "title": "GENA-LM: a family of open-source foundational DNA language models for long sequences" }, "download": { "url": "https://huggingface.co/AIRI-Institute/gena-lm-bert-base-t2t-multi", "label": "Weights", "kind": "weights" }, "source_url": "https://github.com/AIRI-Institute/GENA_LM", "publication": { "first_public_date": "2023-06-13", "first_public_url": "https://doi.org/10.1101/2023.06.12.544594", "first_public_platform": "bioRxiv", "venue": "Nucleic Acids Research", "venue_year": "2025", "venue_url": "https://doi.org/10.1093/nar/gkae1310", "note": "" }, "specifications": { "architecture": "Attention", "parameters": "110M", "input_length": "≈4,500", "tokenizer": "BPE", "pretraining_objective": "MLM", "pretraining_data": "T2T + 1000G SNPs + Multispecies", "input_length_unit": "nt" } }, { "id": "gena-lm-bert-large-t2t", "name": "GENA-BERT-Large-T2T", "category": "Genomic LMs", "evaluation": "FT + LP", "leaderboard_id": "gena-lm-bert-large-t2t", "aliases": [ "Gena-LM-Large" ], "note": "", "paper": { "url": "https://doi.org/10.1093/nar/gkae1310", "label": "Paper", "title": "GENA-LM: a family of open-source foundational DNA language models for long sequences" }, "download": { "url": "https://huggingface.co/AIRI-Institute/gena-lm-bert-large-t2t", "label": "Weights", "kind": "weights" }, "source_url": "https://github.com/AIRI-Institute/GENA_LM", "publication": { "first_public_date": "2023-06-13", "first_public_url": "https://doi.org/10.1101/2023.06.12.544594", "first_public_platform": "bioRxiv", "venue": "Nucleic Acids Research", "venue_year": "2025", "venue_url": "https://doi.org/10.1093/nar/gkae1310", "note": "" }, "specifications": { "architecture": "Attention", "parameters": "336M", "input_length": "≈4,500", "tokenizer": "BPE", "pretraining_objective": "MLM", "pretraining_data": "T2T + 1000G SNPs", "input_length_unit": "nt" } }, { "id": "gena-lm-bigbird-base-t2t", "name": "GENA-LM-BigBird-Base-T2T", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "gena-lm-bigbird-base-t2t", "aliases": [ "Gena-LM-BigBird" ], "note": "", "paper": { "url": "https://doi.org/10.1093/nar/gkae1310", "label": "Paper", "title": "GENA-LM: a family of open-source foundational DNA language models for long sequences" }, "download": { "url": "https://huggingface.co/AIRI-Institute/gena-lm-bigbird-base-t2t", "label": "Weights", "kind": "weights" }, "source_url": "https://github.com/AIRI-Institute/GENA_LM", "publication": { "first_public_date": "2023-06-13", "first_public_url": "https://doi.org/10.1101/2023.06.12.544594", "first_public_platform": "bioRxiv", "venue": "Nucleic Acids Research", "venue_year": "2025", "venue_url": "https://doi.org/10.1093/nar/gkae1310", "note": "" }, "specifications": { "architecture": "Attention", "parameters": "110M", "input_length": "≈36,000", "tokenizer": "BPE", "pretraining_objective": "MLM", "pretraining_data": "T2T + 1000G SNPs", "input_length_unit": "nt" } }, { "id": "nucleotide-transformer-500m-human-ref", "name": "NT-500M-Human-Ref", "category": "Genomic LMs", "evaluation": "FT + LP", "leaderboard_id": "nucleotide-transformer-500m-human-ref", "aliases": [ "NT-500M-Human-Ref" ], "note": "", "paper": { "url": "https://doi.org/10.1038/s41592-024-02523-z", "label": "Paper", "title": "Nucleotide Transformer: building and evaluating robust foundation models for human genomics" }, "download": { "url": "https://huggingface.co/InstaDeepAI/nucleotide-transformer-500m-human-ref", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/InstaDeepAI/nucleotide-transformer-500m-human-ref", "publication": { "first_public_date": "2023-01-15", "first_public_url": "https://doi.org/10.1101/2023.01.11.523679", "first_public_platform": "bioRxiv", "venue": "Nature Methods", "venue_year": "2025", "venue_url": "https://doi.org/10.1038/s41592-024-02523-z", "note": "NT and NTv2 variants share this reference paper; this date does not identify an individual checkpoint release." }, "specifications": { "architecture": "Attention", "parameters": "485M", "input_length": "5,994", "tokenizer": "6-mer", "pretraining_objective": "MLM", "pretraining_data": "Human genome (GRCh38/hg38)", "input_length_unit": "nt" } }, { "id": "nucleotide-transformer-500m-1000g", "name": "NT-500M-1000G", "category": "Genomic LMs", "evaluation": "FT + LP", "leaderboard_id": "nucleotide-transformer-500m-1000g", "aliases": [ "NT-500M-1000g" ], "note": "", "paper": { "url": "https://doi.org/10.1038/s41592-024-02523-z", "label": "Paper", "title": "Nucleotide Transformer: building and evaluating robust foundation models for human genomics" }, "download": { "url": "https://huggingface.co/InstaDeepAI/nucleotide-transformer-500m-1000g", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/InstaDeepAI/nucleotide-transformer-500m-1000g", "publication": { "first_public_date": "2023-01-15", "first_public_url": "https://doi.org/10.1101/2023.01.11.523679", "first_public_platform": "bioRxiv", "venue": "Nature Methods", "venue_year": "2025", "venue_url": "https://doi.org/10.1038/s41592-024-02523-z", "note": "NT and NTv2 variants share this reference paper; this date does not identify an individual checkpoint release." }, "specifications": { "architecture": "Attention", "parameters": "485M", "input_length": "5,994", "tokenizer": "6-mer", "pretraining_objective": "MLM", "pretraining_data": "1000 Genomes Project (3,202 genomes)", "input_length_unit": "nt" } }, { "id": "nucleotide-transformer-2.5b-1000g", "name": "NT-2.5B-1000G", "category": "Genomic LMs", "evaluation": "FT + LP", "leaderboard_id": "nucleotide-transformer-2.5b-1000g", "aliases": [ "NT-2.5B-1000g" ], "note": "", "paper": { "url": "https://doi.org/10.1038/s41592-024-02523-z", "label": "Paper", "title": "Nucleotide Transformer: building and evaluating robust foundation models for human genomics" }, "download": { "url": "https://huggingface.co/InstaDeepAI/nucleotide-transformer-2.5b-1000g", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/InstaDeepAI/nucleotide-transformer-2.5b-1000g", "publication": { "first_public_date": "2023-01-15", "first_public_url": "https://doi.org/10.1101/2023.01.11.523679", "first_public_platform": "bioRxiv", "venue": "Nature Methods", "venue_year": "2025", "venue_url": "https://doi.org/10.1038/s41592-024-02523-z", "note": "NT and NTv2 variants share this reference paper; this date does not identify an individual checkpoint release." }, "specifications": { "architecture": "Attention", "parameters": "2.5B", "input_length": "5,994", "tokenizer": "6-mer", "pretraining_objective": "MLM", "pretraining_data": "1000 Genomes Project (3,202 genomes)", "input_length_unit": "nt" } }, { "id": "nucleotide-transformer-2.5b-multi-species", "name": "NT-2.5B-Multi", "category": "Genomic LMs", "evaluation": "FT + LP", "leaderboard_id": "nucleotide-transformer-2.5b-multi-species", "aliases": [ "NT-2.5B-Multi" ], "note": "", "paper": { "url": "https://doi.org/10.1038/s41592-024-02523-z", "label": "Paper", "title": "Nucleotide Transformer: building and evaluating robust foundation models for human genomics" }, "download": { "url": "https://huggingface.co/InstaDeepAI/nucleotide-transformer-2.5b-multi-species", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/InstaDeepAI/nucleotide-transformer-2.5b-multi-species", "publication": { "first_public_date": "2023-01-15", "first_public_url": "https://doi.org/10.1101/2023.01.11.523679", "first_public_platform": "bioRxiv", "venue": "Nature Methods", "venue_year": "2025", "venue_url": "https://doi.org/10.1038/s41592-024-02523-z", "note": "NT and NTv2 variants share this reference paper; this date does not identify an individual checkpoint release." }, "specifications": { "architecture": "Attention", "parameters": "2.5B", "input_length": "5,994", "tokenizer": "6-mer", "pretraining_objective": "MLM", "pretraining_data": "Multispecies dataset (850 species)", "input_length_unit": "nt" } }, { "id": "nucleotide-transformer-v2-500m-multi-species", "name": "NTv2-500M-Multi", "category": "Genomic LMs", "evaluation": "FT + LP", "leaderboard_id": "nucleotide-transformer-v2-500m-multi-species", "aliases": [ "NTv2-500M-Multi" ], "note": "", "paper": { "url": "https://doi.org/10.1038/s41592-024-02523-z", "label": "Paper", "title": "Nucleotide Transformer: building and evaluating robust foundation models for human genomics" }, "download": { "url": "https://huggingface.co/InstaDeepAI/nucleotide-transformer-v2-500m-multi-species", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/InstaDeepAI/nucleotide-transformer-v2-500m-multi-species", "publication": { "first_public_date": "2023-01-15", "first_public_url": "https://doi.org/10.1101/2023.01.11.523679", "first_public_platform": "bioRxiv", "venue": "Nature Methods", "venue_year": "2025", "venue_url": "https://doi.org/10.1038/s41592-024-02523-z", "note": "NT and NTv2 variants share this reference paper; this date does not identify an individual checkpoint release." }, "specifications": { "architecture": "Attention", "parameters": "498M", "input_length": "12,282", "tokenizer": "6-mer", "pretraining_objective": "MLM", "pretraining_data": "Multispecies dataset (850 species)", "input_length_unit": "nt" } }, { "id": "nucleotide-transformer-v2-250m-multi-species", "name": "NTv2-250M-Multi", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "nucleotide-transformer-v2-250m-multi-species", "aliases": [ "NTv2-250M-Multi" ], "note": "", "paper": { "url": "https://doi.org/10.1038/s41592-024-02523-z", "label": "Paper", "title": "Nucleotide Transformer: building and evaluating robust foundation models for human genomics" }, "download": { "url": "https://huggingface.co/InstaDeepAI/nucleotide-transformer-v2-250m-multi-species", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/InstaDeepAI/nucleotide-transformer-v2-250m-multi-species", "publication": { "first_public_date": "2023-01-15", "first_public_url": "https://doi.org/10.1101/2023.01.11.523679", "first_public_platform": "bioRxiv", "venue": "Nature Methods", "venue_year": "2025", "venue_url": "https://doi.org/10.1038/s41592-024-02523-z", "note": "NT and NTv2 variants share this reference paper; this date does not identify an individual checkpoint release." }, "specifications": { "architecture": "Attention", "parameters": "250M", "input_length": "12,282", "tokenizer": "6-mer", "pretraining_objective": "MLM", "pretraining_data": "Multispecies dataset (850 species)", "input_length_unit": "nt" } }, { "id": "nucleotide-transformer-v2-100m-multi-species", "name": "NTv2-100M-Multi", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "nucleotide-transformer-v2-100m-multi-species", "aliases": [ "NTv2-100M-Multi" ], "note": "", "paper": { "url": "https://doi.org/10.1038/s41592-024-02523-z", "label": "Paper", "title": "Nucleotide Transformer: building and evaluating robust foundation models for human genomics" }, "download": { "url": "https://huggingface.co/InstaDeepAI/nucleotide-transformer-v2-100m-multi-species", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/InstaDeepAI/nucleotide-transformer-v2-100m-multi-species", "publication": { "first_public_date": "2023-01-15", "first_public_url": "https://doi.org/10.1101/2023.01.11.523679", "first_public_platform": "bioRxiv", "venue": "Nature Methods", "venue_year": "2025", "venue_url": "https://doi.org/10.1038/s41592-024-02523-z", "note": "NT and NTv2 variants share this reference paper; this date does not identify an individual checkpoint release." }, "specifications": { "architecture": "Attention", "parameters": "97.9M", "input_length": "12,282", "tokenizer": "6-mer", "pretraining_objective": "MLM", "pretraining_data": "Multispecies dataset (850 species)", "input_length_unit": "nt" } }, { "id": "nucleotide-transformer-v2-50m-multi-species", "name": "NTv2-50M-Multi", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "nucleotide-transformer-v2-50m-multi-species", "aliases": [ "NTv2-50M-Multi" ], "note": "", "paper": { "url": "https://doi.org/10.1038/s41592-024-02523-z", "label": "Paper", "title": "Nucleotide Transformer: building and evaluating robust foundation models for human genomics" }, "download": { "url": "https://huggingface.co/InstaDeepAI/nucleotide-transformer-v2-50m-multi-species", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/InstaDeepAI/nucleotide-transformer-v2-50m-multi-species", "publication": { "first_public_date": "2023-01-15", "first_public_url": "https://doi.org/10.1101/2023.01.11.523679", "first_public_platform": "bioRxiv", "venue": "Nature Methods", "venue_year": "2025", "venue_url": "https://doi.org/10.1038/s41592-024-02523-z", "note": "NT and NTv2 variants share this reference paper; this date does not identify an individual checkpoint release." }, "specifications": { "architecture": "Attention", "parameters": "55.9M", "input_length": "12,282", "tokenizer": "6-mer", "pretraining_objective": "MLM", "pretraining_data": "Multispecies dataset (850 species)", "input_length_unit": "nt" } }, { "id": "ntv3-8m-pre", "name": "NTv3-8M-Pre", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "ntv3-8m-pre", "aliases": [ "NTv3-8M" ], "note": "Pretrained (pre) checkpoint, before supervised post-training. Access requires signing in to Hugging Face and accepting the publisher’s access terms.", "paper": { "url": "https://doi.org/10.64898/2025.12.22.695963", "label": "Paper", "title": "A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction" }, "download": { "url": "https://huggingface.co/InstaDeepAI/NTv3_8M_pre", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/InstaDeepAI/NTv3_8M_pre", "publication": { "first_public_date": "2025-12-25", "first_public_url": "https://doi.org/10.64898/2025.12.22.695963", "first_public_platform": "bioRxiv", "venue": "bioRxiv", "venue_year": "2025", "venue_url": "https://doi.org/10.64898/2025.12.22.695963", "note": "" }, "specifications": { "architecture": "U-Net + Attention", "parameters": "8M", "input_length": "1,000,000", "tokenizer": "One-hot", "pretraining_objective": "MLM", "pretraining_data": "OpenGenome2", "input_length_unit": "nt" } }, { "id": "ntv3-100m-pre", "name": "NTv3-100M-Pre", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "ntv3-100m-pre", "aliases": [ "NTv3-100M" ], "note": "Pretrained (pre) checkpoint, before supervised post-training. Access requires signing in to Hugging Face and accepting the publisher’s access terms.", "paper": { "url": "https://doi.org/10.64898/2025.12.22.695963", "label": "Paper", "title": "A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction" }, "download": { "url": "https://huggingface.co/InstaDeepAI/NTv3_100M_pre", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/InstaDeepAI/NTv3_100M_pre", "publication": { "first_public_date": "2025-12-25", "first_public_url": "https://doi.org/10.64898/2025.12.22.695963", "first_public_platform": "bioRxiv", "venue": "bioRxiv", "venue_year": "2025", "venue_url": "https://doi.org/10.64898/2025.12.22.695963", "note": "" }, "specifications": { "architecture": "U-Net + Attention", "parameters": "100M", "input_length": "1,000,000", "tokenizer": "One-hot", "pretraining_objective": "MLM", "pretraining_data": "OpenGenome2", "input_length_unit": "nt" } }, { "id": "ntv3-650m-pre", "name": "NTv3-650M-Pre", "category": "Genomic LMs", "evaluation": "FT + LP", "leaderboard_id": "ntv3-650m-pre", "aliases": [ "NTv3-650M" ], "note": "Pretrained (pre) checkpoint, before supervised post-training. Access requires signing in to Hugging Face and accepting the publisher’s access terms.", "paper": { "url": "https://doi.org/10.64898/2025.12.22.695963", "label": "Paper", "title": "A foundational model for joint sequence-function multi-species modeling at scale for long-range genomic prediction" }, "download": { "url": "https://huggingface.co/InstaDeepAI/NTv3_650M_pre", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/InstaDeepAI/NTv3_650M_pre", "publication": { "first_public_date": "2025-12-25", "first_public_url": "https://doi.org/10.64898/2025.12.22.695963", "first_public_platform": "bioRxiv", "venue": "bioRxiv", "venue_year": "2025", "venue_url": "https://doi.org/10.64898/2025.12.22.695963", "note": "" }, "specifications": { "architecture": "U-Net + Attention", "parameters": "650M", "input_length": "1,000,000", "tokenizer": "One-hot", "pretraining_objective": "MLM", "pretraining_data": "OpenGenome2", "input_length_unit": "nt" } }, { "id": "hyena_large_1m", "name": "Hyena-Large", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "hyena_large_1m", "aliases": [ "Hyena-Large" ], "note": "", "paper": { "url": "https://arxiv.org/abs/2306.15794", "label": "Paper", "title": "Hyenadna: Long-range genomic sequence modeling at single nucleotide resolution" }, "download": { "url": "https://huggingface.co/LongSafari/hyenadna-large-1m-seqlen", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/LongSafari/hyenadna-large-1m-seqlen", "publication": { "first_public_date": "2023-06-27", "first_public_url": "https://arxiv.org/abs/2306.15794v1", "first_public_platform": "arXiv", "venue": "NeurIPS", "venue_year": "2023", "venue_url": "https://arxiv.org/abs/2306.15794", "note": "" }, "specifications": { "architecture": "Hyena", "parameters": "6.6M", "input_length": "1,000,000", "tokenizer": "Char", "pretraining_objective": "NTP", "pretraining_data": "Human genome (GRCh38/hg38)", "input_length_unit": "nt" } }, { "id": "hyena_medium_160k", "name": "Hyena-Medium", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "hyena_medium_160k", "aliases": [ "Hyena-Medium" ], "note": "", "paper": { "url": "https://arxiv.org/abs/2306.15794", "label": "Paper", "title": "Hyenadna: Long-range genomic sequence modeling at single nucleotide resolution" }, "download": { "url": "https://huggingface.co/LongSafari/hyenadna-medium-160k-seqlen", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/LongSafari/hyenadna-medium-160k-seqlen", "publication": { "first_public_date": "2023-06-27", "first_public_url": "https://arxiv.org/abs/2306.15794v1", "first_public_platform": "arXiv", "venue": "NeurIPS", "venue_year": "2023", "venue_url": "https://arxiv.org/abs/2306.15794", "note": "" }, "specifications": { "architecture": "Hyena", "parameters": "3.3M", "input_length": "160,000", "tokenizer": "Char", "pretraining_objective": "NTP", "pretraining_data": "Human genome (GRCh38/hg38)", "input_length_unit": "nt" } }, { "id": "caduceus_ph_seqlen_131k_d_model_256", "name": "Caduceus-PH", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "caduceus_ph_seqlen_131k_d_model_256", "aliases": [ "Caduceus-Ph" ], "note": "", "paper": { "url": "https://arxiv.org/abs/2403.03234", "label": "Paper", "title": "Caduceus: Bi-Directional Equivariant Long-Range DNA Sequence Modeling" }, "download": { "url": "https://huggingface.co/kuleshov-group/caduceus-ph_seqlen-131k_d_model-256_n_layer-16", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/kuleshov-group/caduceus-ph_seqlen-131k_d_model-256_n_layer-16", "publication": { "first_public_date": "2024-03-05", "first_public_url": "https://arxiv.org/abs/2403.03234v1", "first_public_platform": "arXiv", "venue": "ICML", "venue_year": "2024", "venue_url": "https://arxiv.org/abs/2403.03234", "note": "" }, "specifications": { "architecture": "SSM", "parameters": "7.73M", "input_length": "131,072", "tokenizer": "Char", "pretraining_objective": "MLM", "pretraining_data": "Human genome (GRCh37/hg19)", "input_length_unit": "nt" } }, { "id": "caduceus_ps_seqlen_131k_d_model_256", "name": "Caduceus-PS", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "caduceus_ps_seqlen_131k_d_model_256", "aliases": [ "Caduceus-Ps" ], "note": "", "paper": { "url": "https://arxiv.org/abs/2403.03234", "label": "Paper", "title": "Caduceus: Bi-Directional Equivariant Long-Range DNA Sequence Modeling" }, "download": { "url": "https://huggingface.co/kuleshov-group/caduceus-ps_seqlen-131k_d_model-256_n_layer-16", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/kuleshov-group/caduceus-ps_seqlen-131k_d_model-256_n_layer-16", "publication": { "first_public_date": "2024-03-05", "first_public_url": "https://arxiv.org/abs/2403.03234v1", "first_public_platform": "arXiv", "venue": "ICML", "venue_year": "2024", "venue_url": "https://arxiv.org/abs/2403.03234", "note": "" }, "specifications": { "architecture": "SSM", "parameters": "7.73M", "input_length": "131,072", "tokenizer": "Char", "pretraining_objective": "MLM", "pretraining_data": "Human genome (GRCh37/hg19)", "input_length_unit": "nt" } }, { "id": "generator_eukaryote_1_2b", "name": "GENERator-1.2B", "category": "Genomic LMs", "evaluation": "FT + LP", "leaderboard_id": "generator_eukaryote_1_2b", "aliases": [ "Generator-1.2B" ], "note": "", "paper": { "url": "https://arxiv.org/abs/2502.07272", "label": "Paper", "title": "GENERator: A Long-Context Generative Genomic Foundation Model" }, "download": { "url": "https://huggingface.co/GenerTeam/GENERator-eukaryote-1.2b-base", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/GenerTeam/GENERator-eukaryote-1.2b-base", "publication": { "first_public_date": "2025-02-11", "first_public_url": "https://arxiv.org/abs/2502.07272v1", "first_public_platform": "arXiv", "venue": "arXiv", "venue_year": "2025", "venue_url": "https://arxiv.org/abs/2502.07272", "note": "" }, "specifications": { "architecture": "Attention", "parameters": "1.2B", "input_length": "98,000", "tokenizer": "6-mer", "pretraining_objective": "NTP", "pretraining_data": "RefSeq eukaryotic genomes", "input_length_unit": "nt" } }, { "id": "grover", "name": "GROVER", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "grover", "aliases": [ "Grover" ], "note": "", "paper": { "url": "https://doi.org/10.1038/s42256-024-00872-0", "label": "Paper", "title": "DNA language model GROVER learns sequence context in the human genome" }, "download": { "url": "https://huggingface.co/PoetschLab/GROVER", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/PoetschLab/GROVER", "publication": { "first_public_date": "2023-07-19", "first_public_url": "https://doi.org/10.1101/2023.07.19.549677", "first_public_platform": "bioRxiv", "venue": "Nature Machine Intelligence", "venue_year": "2024", "venue_url": "https://doi.org/10.1038/s42256-024-00872-0", "note": "" }, "specifications": { "architecture": "Attention", "parameters": "86.5M", "input_length": "Variable", "tokenizer": "BPE", "pretraining_objective": "MLM", "pretraining_data": "Human genome (GRCh37/hg19)", "input_length_unit": "nt" } }, { "id": "phylogpn", "name": "PhyloGPN", "category": "Genomic LMs", "evaluation": "FT", "leaderboard_id": "phylogpn", "aliases": [ "PhyloGPN" ], "note": "", "paper": { "url": "https://arxiv.org/abs/2503.03773v2", "label": "Paper", "title": "A Phylogenetic Approach to Genomic Language Modeling" }, "download": { "url": "https://huggingface.co/songlab/PhyloGPN", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/songlab/PhyloGPN", "publication": { "first_public_date": "2025-03-04", "first_public_url": "https://arxiv.org/abs/2503.03773v1", "first_public_platform": "arXiv", "venue": "arXiv", "venue_year": "2025", "venue_url": "https://arxiv.org/abs/2503.03773v2", "note": "" }, "specifications": { "architecture": "CNN", "parameters": "83.2M", "input_length": "Variable", "tokenizer": "Char", "pretraining_objective": "Custom objective", "pretraining_data": "Zoonomia (447 mammalian genomes)", "input_length_unit": "nt" } }, { "id": "evo2_1b_base", "name": "Evo2-1B-Base", "category": "Genomic LMs", "evaluation": "FT + LP", "leaderboard_id": "evo2_1b_base", "aliases": [ "Evo2-1B-Base" ], "note": "", "paper": { "url": "https://doi.org/10.1038/s41586-026-10176-5", "label": "Paper", "title": "Genome modelling and design across all domains of life with Evo 2" }, "download": { "url": "https://huggingface.co/arcinstitute/evo2_1b_base", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/arcinstitute/evo2_1b_base", "publication": { "first_public_date": "2025-02-21", "first_public_url": "https://doi.org/10.1101/2025.02.18.638918", "first_public_platform": "bioRxiv", "venue": "Nature", "venue_year": "2026", "venue_url": "https://doi.org/10.1038/s41586-026-10176-5", "note": "" }, "specifications": { "architecture": "Hyena + Attention", "parameters": "1.1B", "input_length": "8,192", "tokenizer": "Char", "pretraining_objective": "NTP", "pretraining_data": "OpenGenome2", "input_length_unit": "nt" } }, { "id": "evo2-7b", "name": "Evo2-7B", "category": "Genomic LMs", "evaluation": "LP", "leaderboard_id": null, "aliases": [], "note": "Evaluated by linear probing; outside the 43-model main fine-tuning leaderboard.", "paper": { "url": "https://doi.org/10.1038/s41586-026-10176-5", "label": "Paper", "title": "Genome modelling and design across all domains of life with Evo 2" }, "download": { "url": "https://huggingface.co/arcinstitute/evo2_7b", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/arcinstitute/evo2_7b", "publication": { "first_public_date": "2025-02-21", "first_public_url": "https://doi.org/10.1101/2025.02.18.638918", "first_public_platform": "bioRxiv", "venue": "Nature", "venue_year": "2026", "venue_url": "https://doi.org/10.1038/s41586-026-10176-5", "note": "" }, "specifications": { "architecture": "Hyena + Attention", "parameters": "7B", "input_length": "1,048,576", "tokenizer": "Char", "pretraining_objective": "NTP", "pretraining_data": "OpenGenome2", "input_length_unit": "nt" } }, { "id": "evo2-40b", "name": "Evo2-40B", "category": "Genomic LMs", "evaluation": "LP", "leaderboard_id": null, "aliases": [], "note": "Evaluated by linear probing; outside the 43-model main fine-tuning leaderboard.", "paper": { "url": "https://doi.org/10.1038/s41586-026-10176-5", "label": "Paper", "title": "Genome modelling and design across all domains of life with Evo 2" }, "download": { "url": "https://huggingface.co/arcinstitute/evo2_40b", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/arcinstitute/evo2_40b", "publication": { "first_public_date": "2025-02-21", "first_public_url": "https://doi.org/10.1101/2025.02.18.638918", "first_public_platform": "bioRxiv", "venue": "Nature", "venue_year": "2026", "venue_url": "https://doi.org/10.1038/s41586-026-10176-5", "note": "" }, "specifications": { "architecture": "Hyena + Attention", "parameters": "40B", "input_length": "1,048,576", "tokenizer": "Char", "pretraining_objective": "NTP", "pretraining_data": "OpenGenome2", "input_length_unit": "nt" } }, { "id": "rnafm", "name": "RNAFM", "category": "RNA LMs", "evaluation": "FT", "leaderboard_id": "rnafm", "aliases": [ "RNAFM" ], "note": "", "paper": { "url": "https://arxiv.org/abs/2204.00300", "label": "Paper", "title": "Interpretable rna foundation model from unannotated data for highly accurate rna structure and function predictions" }, "download": { "url": "https://huggingface.co/cuhkaih/rnafm", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/cuhkaih/rnafm", "publication": { "first_public_date": "2022-04-01", "first_public_url": "https://arxiv.org/abs/2204.00300v1", "first_public_platform": "arXiv", "venue": "arXiv", "venue_year": "2022", "venue_url": "https://arxiv.org/abs/2204.00300", "note": "" }, "specifications": { "architecture": "Attention", "parameters": "99.5M", "input_length": "1,024", "tokenizer": "Char", "pretraining_objective": "MLM", "pretraining_data": "RNAcentral", "input_length_unit": "nt" } }, { "id": "rnaernie", "name": "RNAErnie", "category": "RNA LMs", "evaluation": "FT", "leaderboard_id": "rnaernie", "aliases": [ "RNAErnie" ], "note": "Original RNAErnie release and download instructions; the newer RNAErnie2 model is a separate release.", "paper": { "url": "https://doi.org/10.1038/s42256-024-00836-4", "label": "Paper", "title": "Multi-purpose RNA language modelling with motif-aware pretraining and type-guided fine-tuning" }, "download": { "url": "https://github.com/CatIIIIIIII/RNAErnie#3-download-pre-trained-models", "label": "Weight downloads", "kind": "weights" }, "source_url": "https://github.com/CatIIIIIIII/RNAErnie", "publication": { "first_public_date": "2024-05-13", "first_public_url": "https://doi.org/10.1038/s42256-024-00836-4", "first_public_platform": "Nature Machine Intelligence", "venue": "Nature Machine Intelligence", "venue_year": "2024", "venue_url": "https://doi.org/10.1038/s42256-024-00836-4", "note": "Publisher's first online publication date." }, "specifications": { "architecture": "Attention", "parameters": "105M", "input_length": "512", "tokenizer": "Char", "pretraining_objective": "MLM", "pretraining_data": "RNAcentral", "input_length_unit": "nt" } }, { "id": "splicebert", "name": "SpliceBERT", "category": "RNA LMs", "evaluation": "FT", "leaderboard_id": "splicebert", "aliases": [ "SpliceBERT" ], "note": "Model archive includes multiple variants; use the vertebrate-pretrained model described in the paper.", "paper": { "url": "https://doi.org/10.1093/bib/bbae163", "label": "Paper", "title": "Self-supervised learning on millions of primary RNA sequences from 72 vertebrates improves sequence-based RNA splicing prediction" }, "download": { "url": "https://zenodo.org/records/7995778", "label": "Weights", "kind": "weights" }, "source_url": "https://github.com/chenkenbio/SpliceBERT", "publication": { "first_public_date": "2023-02-03", "first_public_url": "https://doi.org/10.1101/2023.01.31.526427", "first_public_platform": "bioRxiv", "venue": "Briefings in Bioinformatics", "venue_year": "2024", "venue_url": "https://doi.org/10.1093/bib/bbae163", "note": "" }, "specifications": { "architecture": "Attention", "parameters": "19.4M", "input_length": "1,024", "tokenizer": "Char", "pretraining_objective": "MLM", "pretraining_data": "72 vertebrates RNA sequences", "input_length_unit": "nt" } }, { "id": "rinalmo", "name": "RiNALMo", "category": "RNA LMs", "evaluation": "FT", "leaderboard_id": "rinalmo", "aliases": [ "RiNALMo" ], "note": "Use rinalmo_giga_pretrained.pt for the 650M model.", "paper": { "url": "https://doi.org/10.1038/s41467-025-60872-5", "label": "Paper", "title": "Rinalmo: General-purpose rna language models can generalize well on structure prediction tasks" }, "download": { "url": "https://zenodo.org/records/15043668", "label": "Weights", "kind": "weights" }, "source_url": "https://github.com/lbcb-sci/RiNALMo", "publication": { "first_public_date": "2024-02-29", "first_public_url": "https://arxiv.org/abs/2403.00043v1", "first_public_platform": "arXiv", "venue": "Nature Communications", "venue_year": "2025", "venue_url": "https://doi.org/10.1038/s41467-025-60872-5", "note": "" }, "specifications": { "architecture": "Attention", "parameters": "650M", "input_length": "8,192", "tokenizer": "Char", "pretraining_objective": "MLM", "pretraining_data": "Public ncRNA", "input_length_unit": "nt" } }, { "id": "orthrus-4track", "name": "Orthrus-4-track", "category": "RNA LMs", "evaluation": "FT", "leaderboard_id": "orthrus-4track", "aliases": [ "Orthrus-4Track" ], "note": "", "paper": { "url": "https://doi.org/10.1038/s41592-026-03064-3", "label": "Paper", "title": "Orthrus: toward evolutionary and functional RNA foundation models" }, "download": { "url": "https://huggingface.co/antichronology/orthrus-4-track", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/antichronology/orthrus-4-track", "publication": { "first_public_date": "2024-10-12", "first_public_url": "https://doi.org/10.1101/2024.10.10.617658", "first_public_platform": "bioRxiv", "venue": "Nature Methods", "venue_year": "2026", "venue_url": "https://doi.org/10.1038/s41592-026-03064-3", "note": "" }, "specifications": { "architecture": "SSM", "parameters": "10M", "input_length": "Variable", "tokenizer": "One-hot", "pretraining_objective": "Contrastive", "pretraining_data": "45M+ mature mRNA transcripts", "input_length_unit": "nt" } }, { "id": "rnabert", "name": "RNA-BERT", "category": "RNA LMs", "evaluation": "FT", "leaderboard_id": "rnabert", "aliases": [ "RNA-BERT" ], "note": "", "paper": { "url": "https://doi.org/10.1093/nargab/lqac012", "label": "Paper", "title": "Informative RNA base embedding for RNA structural alignment and clustering by deep representation learning" }, "download": { "url": "https://drive.google.com/file/d/1sT6jlv9vrpX0npKmnbFeOqZ1JZDrZTQ2/view", "label": "Weights", "kind": "weights" }, "source_url": "https://github.com/mana438/RNABERT", "publication": { "first_public_date": "2021-08-24", "first_public_url": "https://doi.org/10.1101/2021.08.23.457433", "first_public_platform": "bioRxiv", "venue": "NAR Genomics and Bioinformatics", "venue_year": "2022", "venue_url": "https://doi.org/10.1093/nargab/lqac012", "note": "" }, "specifications": { "architecture": "Attention", "parameters": "0.51M", "input_length": "440", "tokenizer": "Char", "pretraining_objective": "MLM + SAL", "pretraining_data": "RNAcentral + Rfam", "input_length_unit": "nt" } }, { "id": "esm-1b", "name": "ESM-1b", "category": "Protein LMs", "evaluation": "FT", "leaderboard_id": null, "aliases": [], "note": "Official esm1b_t33_650M_UR50S checkpoint (direct .pt download).", "paper": { "url": "https://doi.org/10.1073/pnas.2016239118", "label": "Paper", "title": "Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences" }, "download": { "url": "https://dl.fbaipublicfiles.com/fair-esm/models/esm1b_t33_650M_UR50S.pt", "label": "Weights", "kind": "weights" }, "source_url": "https://github.com/facebookresearch/esm", "specifications": { "architecture": "Attention", "parameters": "650M", "input_length": "1,024", "tokenizer": "Char", "pretraining_objective": "MLM", "pretraining_data": "UR50/S", "input_length_unit": "aa" } }, { "id": "esm-2-150m", "name": "ESM-2-150M", "category": "Protein LMs", "evaluation": "FT", "leaderboard_id": null, "aliases": [], "note": "Official esm2_t30_150M_UR50D checkpoint (direct .pt download).", "paper": { "url": "https://doi.org/10.1126/science.ade2574", "label": "Paper", "title": "Evolutionary-scale prediction of atomic-level protein structure with a language model" }, "download": { "url": "https://dl.fbaipublicfiles.com/fair-esm/models/esm2_t30_150M_UR50D.pt", "label": "Weights", "kind": "weights" }, "source_url": "https://github.com/facebookresearch/esm", "specifications": { "architecture": "Attention", "parameters": "150M", "input_length": "1,024", "tokenizer": "Char", "pretraining_objective": "MLM", "pretraining_data": "UR50/D", "input_length_unit": "aa" } }, { "id": "esm-2-650m", "name": "ESM-2-650M", "category": "Protein LMs", "evaluation": "FT", "leaderboard_id": null, "aliases": [], "note": "Official esm2_t33_650M_UR50D checkpoint (direct .pt download).", "paper": { "url": "https://doi.org/10.1126/science.ade2574", "label": "Paper", "title": "Evolutionary-scale prediction of atomic-level protein structure with a language model" }, "download": { "url": "https://dl.fbaipublicfiles.com/fair-esm/models/esm2_t33_650M_UR50D.pt", "label": "Weights", "kind": "weights" }, "source_url": "https://github.com/facebookresearch/esm", "specifications": { "architecture": "Attention", "parameters": "650M", "input_length": "1,024", "tokenizer": "Char", "pretraining_objective": "MLM", "pretraining_data": "UR50/D", "input_length_unit": "aa" } }, { "id": "esm-2-3b", "name": "ESM-2-3B", "category": "Protein LMs", "evaluation": "FT", "leaderboard_id": null, "aliases": [], "note": "Official esm2_t36_3B_UR50D checkpoint (direct .pt download).", "paper": { "url": "https://doi.org/10.1126/science.ade2574", "label": "Paper", "title": "Evolutionary-scale prediction of atomic-level protein structure with a language model" }, "download": { "url": "https://dl.fbaipublicfiles.com/fair-esm/models/esm2_t36_3B_UR50D.pt", "label": "Weights", "kind": "weights" }, "source_url": "https://github.com/facebookresearch/esm", "specifications": { "architecture": "Attention", "parameters": "3B", "input_length": "1,024", "tokenizer": "Char", "pretraining_objective": "MLM", "pretraining_data": "UR50/D", "input_length_unit": "aa" } }, { "id": "esm-2-15b", "name": "ESM-2-15B", "category": "Protein LMs", "evaluation": "LP", "leaderboard_id": null, "aliases": [], "note": "Evaluated by linear probing on protein tasks.", "paper": { "url": "https://doi.org/10.1126/science.ade2574", "label": "Paper", "title": "Evolutionary-scale prediction of atomic-level protein structure with a language model" }, "download": { "url": "https://huggingface.co/facebook/esm2_t48_15B_UR50D", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/facebook/esm2_t48_15B_UR50D", "specifications": { "architecture": "Attention", "parameters": "15B", "input_length": "1,024", "tokenizer": "Char", "pretraining_objective": "MLM", "pretraining_data": "UR50/D", "input_length_unit": "aa" } }, { "id": "esm-c-300m", "name": "ESM-C-300M", "category": "Protein LMs", "evaluation": "FT", "leaderboard_id": null, "aliases": [], "note": "2024-12 checkpoint used by the benchmark.", "paper": { "url": "https://doi.org/10.64898/2026.06.03.729735", "label": "Paper", "title": "Language Modeling Materializes a World Model of Protein Biology" }, "download": { "url": "https://huggingface.co/EvolutionaryScale/esmc-300m-2024-12", "label": "Weights", "kind": "weights" }, "source_url": "https://github.com/evolutionaryscale/esm", "specifications": { "architecture": "Attention", "parameters": "300M", "input_length": "2,048", "tokenizer": "Char", "pretraining_objective": "MLM", "pretraining_data": "UniRef + MGnify + JGI", "input_length_unit": "aa" } }, { "id": "esm-c-600m", "name": "ESM-C-600M", "category": "Protein LMs", "evaluation": "FT", "leaderboard_id": null, "aliases": [], "note": "2024-12 checkpoint used by the benchmark.", "paper": { "url": "https://doi.org/10.64898/2026.06.03.729735", "label": "Paper", "title": "Language Modeling Materializes a World Model of Protein Biology" }, "download": { "url": "https://huggingface.co/EvolutionaryScale/esmc-600m-2024-12", "label": "Weights", "kind": "weights" }, "source_url": "https://github.com/evolutionaryscale/esm", "specifications": { "architecture": "Attention", "parameters": "600M", "input_length": "2,048", "tokenizer": "Char", "pretraining_objective": "MLM", "pretraining_data": "UniRef + MGnify + JGI", "input_length_unit": "aa" } }, { "id": "esm-c-6b", "name": "ESM-C-6B", "category": "Protein LMs", "evaluation": "FT", "leaderboard_id": null, "aliases": [], "note": "", "paper": { "url": "https://doi.org/10.64898/2026.06.03.729735", "label": "Paper", "title": "Language Modeling Materializes a World Model of Protein Biology" }, "download": { "url": "https://huggingface.co/biohub/ESMC-6B", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/biohub/ESMC-6B", "specifications": { "architecture": "Attention", "parameters": "6B", "input_length": "2,048", "tokenizer": "Char", "pretraining_objective": "MLM", "pretraining_data": "UniRef + MGnify + JGI", "input_length_unit": "aa" } }, { "id": "esm-3-small-open", "name": "ESM-3-small-open", "category": "Protein LMs", "evaluation": "FT", "leaderboard_id": null, "aliases": [], "note": "", "paper": { "url": "https://doi.org/10.1126/science.ads0018", "label": "Paper", "title": "Simulating 500 million years of evolution with a language model" }, "download": { "url": "https://huggingface.co/biohub/esm3-sm-open-v1", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/biohub/esm3-sm-open-v1", "specifications": { "architecture": "Attention", "parameters": "1.4B", "input_length": "2,048", "tokenizer": "Char", "pretraining_objective": "MLM", "pretraining_data": "Sequence + structure + function data", "input_length_unit": "aa" } }, { "id": "xtrimopglm-1b", "name": "xTrimoPGLM-1B", "category": "Protein LMs", "evaluation": "FT", "leaderboard_id": null, "aliases": [], "note": "", "paper": { "url": "https://doi.org/10.1038/s41592-025-02636-z", "label": "Paper", "title": "Xtrimopglm: unified 100-billion-parameter pretrained transformer for deciphering the language of proteins" }, "download": { "url": "https://huggingface.co/biomap-research/xtrimopglm-1b-mlm", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/biomap-research/xtrimopglm-1b-mlm", "specifications": { "architecture": "Attention", "parameters": "1B", "input_length": "2,048", "tokenizer": "Char", "pretraining_objective": "MLM", "pretraining_data": "UniRef90 + ColabFoldDB", "input_length_unit": "aa" } }, { "id": "xtrimopglm-3b", "name": "xTrimoPGLM-3B", "category": "Protein LMs", "evaluation": "FT", "leaderboard_id": null, "aliases": [], "note": "", "paper": { "url": "https://doi.org/10.1038/s41592-025-02636-z", "label": "Paper", "title": "Xtrimopglm: unified 100-billion-parameter pretrained transformer for deciphering the language of proteins" }, "download": { "url": "https://huggingface.co/biomap-research/xtrimopglm-3b-mlm", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/biomap-research/xtrimopglm-3b-mlm", "specifications": { "architecture": "Attention", "parameters": "3B", "input_length": "2,048", "tokenizer": "Char", "pretraining_objective": "MLM", "pretraining_data": "UniRef90 + ColabFoldDB", "input_length_unit": "aa" } }, { "id": "enformer", "name": "Enformer", "category": "Task-specific", "evaluation": "FT", "leaderboard_id": "enformer", "aliases": [ "Enformer" ], "note": "", "paper": { "url": "https://doi.org/10.1038/s41592-021-01252-x", "label": "Paper", "title": "Effective gene expression prediction from sequence by integrating long-range interactions" }, "download": { "url": "https://huggingface.co/EleutherAI/enformer-official-rough", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/EleutherAI/enformer-official-rough", "publication": { "first_public_date": "2021-04-08", "first_public_url": "https://doi.org/10.1101/2021.04.07.438649", "first_public_platform": "bioRxiv", "venue": "Nature Methods", "venue_year": "2021", "venue_url": "https://doi.org/10.1038/s41592-021-01252-x", "note": "" }, "specifications": { "architecture": "CNN + Attention", "parameters": "251M", "input_length": "196,608", "tokenizer": "One-hot", "pretraining_objective": "Supervised", "pretraining_data": "Human genome (5,313 tracks) + Mouse genome (1,643 tracks)", "input_length_unit": "nt" } }, { "id": "borzoi", "name": "Borzoi", "category": "Task-specific", "evaluation": "FT", "leaderboard_id": "borzoi", "aliases": [ "Borzoi" ], "note": "", "paper": { "url": "https://doi.org/10.1038/s41588-024-02053-6", "label": "Paper", "title": "Predicting RNA-seq coverage from DNA sequence as a unifying model of gene regulation" }, "download": { "url": "https://huggingface.co/johahi/borzoi-replicate-0", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/johahi/borzoi-replicate-0", "publication": { "first_public_date": "2023-09-01", "first_public_url": "https://doi.org/10.1101/2023.08.30.555582", "first_public_platform": "bioRxiv", "venue": "Nature Genetics", "venue_year": "2025", "venue_url": "https://doi.org/10.1038/s41588-024-02053-6", "note": "" }, "specifications": { "architecture": "CNN + Attention", "parameters": "185M", "input_length": "524,288", "tokenizer": "One-hot", "pretraining_objective": "Supervised", "pretraining_data": "Human genome (7,611 tracks) + Mouse genome (2,608 tracks)", "input_length_unit": "nt" } }, { "id": "alphagenome", "name": "AlphaGenome", "category": "Task-specific", "evaluation": "FT + LP", "leaderboard_id": "alphagenome", "aliases": [ "AlphaGenome" ], "note": "Access requires signing in to Hugging Face and accepting the publisher’s access terms.", "paper": { "url": "https://doi.org/10.1038/s41586-025-10014-0", "label": "Paper", "title": "Advancing regulatory variant effect prediction with AlphaGenome" }, "download": { "url": "https://huggingface.co/google/alphagenome-all-folds", "label": "Weights", "kind": "weights" }, "source_url": "https://huggingface.co/google/alphagenome-all-folds", "publication": { "first_public_date": "2025-06-27", "first_public_url": "https://doi.org/10.1101/2025.06.25.661532", "first_public_platform": "bioRxiv", "venue": "Nature", "venue_year": "2026", "venue_url": "https://doi.org/10.1038/s41586-025-10014-0", "note": "" }, "specifications": { "architecture": "U-Net + Attention", "parameters": "450M", "input_length": "1,048,576", "tokenizer": "One-hot", "pretraining_objective": "Supervised + Distillation", "pretraining_data": "Human & Mouse (ENCODE/GTEx)", "input_length_unit": "nt" } }, { "id": "deepsea", "name": "DeepSEA", "category": "Task-specific", "evaluation": "FT", "leaderboard_id": null, "aliases": [], "note": "", "paper": { "url": "https://doi.org/10.1038/nmeth.3547", "label": "Paper", "title": "Predicting effects of noncoding variants with deep learning--based sequence model" }, "download": { "url": "https://zenodo.org/record/1466993", "label": "Weights", "kind": "weights" }, "source_url": "https://zenodo.org/record/1466993", "specifications": { "architecture": "CNN", "parameters": "52.8M", "input_length": "1,000", "tokenizer": "One-hot", "pretraining_objective": "Supervised", "pretraining_data": "ENCODE + Roadmap Epigenomics", "input_length_unit": "nt" } }, { "id": "spliceai", "name": "SpliceAI", "category": "Task-specific", "evaluation": "Fixed", "leaderboard_id": null, "aliases": [], "note": "", "paper": { "url": "https://doi.org/10.1016/j.cell.2018.12.015", "label": "Paper", "title": "Predicting splicing from primary sequence with deep learning" }, "download": { "url": "https://github.com/Illumina/SpliceAI", "label": "Weights & code", "kind": "weights" }, "source_url": "https://github.com/Illumina/SpliceAI", "specifications": { "architecture": "CNN", "parameters": "3.48M", "input_length": "Variable", "tokenizer": "One-hot", "pretraining_objective": "Supervised", "pretraining_data": "Human reference transcript splice junctions", "input_length_unit": "nt" } }, { "id": "alphamissense", "name": "AlphaMissense", "category": "Task-specific", "evaluation": "Fixed", "leaderboard_id": null, "aliases": [], "note": "Official precomputed predictions. Trained model weights are not released. The Google Cloud console requests sign-in to browse the release.", "paper": { "url": "https://doi.org/10.1126/science.adg7492", "label": "Paper", "title": "Accurate proteome-wide missense variant effect prediction with AlphaMissense" }, "download": { "url": "https://console.cloud.google.com/storage/browser/dm_alphamissense", "label": "Prediction scores", "kind": "scores" }, "source_url": "https://github.com/google-deepmind/alphamissense", "specifications": { "architecture": "Deep neural network", "parameters": "—", "input_length": "—", "tokenizer": "—", "pretraining_objective": "—", "pretraining_data": "Human/primate variants", "input_length_unit": "nt" } }, { "id": "targetfinder", "name": "TargetFinder", "category": "Task-specific", "evaluation": "Scratch", "leaderboard_id": null, "aliases": [], "note": "Retrained on benchmark splits; the source provides the method implementation.", "paper": { "url": "https://doi.org/10.1038/ng.3539", "label": "Paper", "title": "Enhancer--promoter interactions are encoded by complex genomic signatures on looping chromatin" }, "download": { "url": "https://github.com/shwhalen/targetfinder", "label": "Code", "kind": "code" }, "source_url": "https://github.com/shwhalen/targetfinder", "specifications": { "architecture": "Gradient-boosted trees", "parameters": "—", "input_length": "—", "tokenizer": "—", "pretraining_objective": "—", "pretraining_data": "—", "input_length_unit": "nt" } }, { "id": "saluki", "name": "Saluki", "category": "Task-specific", "evaluation": "Fixed", "leaderboard_id": null, "aliases": [], "note": "Human output of the released f1_c1 checkpoint.", "paper": { "url": "https://doi.org/10.1186/s13059-022-02811-x", "label": "Paper", "title": "The genetic and biochemical determinants of mRNA degradation rates in mammals" }, "download": { "url": "https://zenodo.org/records/6326409", "label": "Weights & data", "kind": "weights" }, "source_url": "https://github.com/vagarwal87/saluki_paper", "specifications": { "architecture": "CNN + GRU", "parameters": "0.16M", "input_length": "12,288", "tokenizer": "One-hot", "pretraining_objective": "Supervised", "pretraining_data": "Human + mouse mRNA half-life", "input_length_unit": "nt" } }, { "id": "ribonn", "name": "RiboNN", "category": "Task-specific", "evaluation": "Fixed", "leaderboard_id": null, "aliases": [], "note": "Human checkpoint: outer fold 0, inner fold 0.", "paper": { "url": "https://doi.org/10.1038/s41587-025-02712-x", "label": "Paper", "title": "Predicting the translation efficiency of messenger RNA in mammalian cells" }, "download": { "url": "https://zenodo.org/records/17258709", "label": "Weights", "kind": "weights" }, "source_url": "https://github.com/Sanofi-Public/RiboNN", "specifications": { "architecture": "CNN", "parameters": "≈0.24M", "input_length": "13,318", "tokenizer": "One-hot", "pretraining_objective": "Supervised", "pretraining_data": "Human + mouse Ribo-seq/RNA-seq", "input_length_unit": "nt" } }, { "id": "phylop", "name": "phyloP", "category": "Task-specific", "evaluation": "Fixed", "leaderboard_id": null, "aliases": [], "note": "Conservation software and scores; this method has no neural-network checkpoint.", "paper": { "url": "https://doi.org/10.1101/gr.097857.109", "label": "Paper", "title": "Detection of nonneutral substitution rates on mammalian phylogenies" }, "download": { "url": "https://github.com/CshlSiepelLab/phast", "label": "Software", "kind": "software" }, "source_url": "https://github.com/CshlSiepelLab/phast", "specifications": { "architecture": "Phylogenetic model", "parameters": "—", "input_length": "—", "tokenizer": "—", "pretraining_objective": "—", "pretraining_data": "—", "input_length_unit": "nt" } }, { "id": "baseline_cnn", "name": "CNN", "category": "Baselines", "evaluation": "Scratch", "leaderboard_id": "baseline_cnn", "aliases": [ "CNN" ], "note": "Trained from scratch in this benchmark; no pretrained checkpoint. Paper describes the referenced baseline framework.", "paper": { "url": "https://doi.org/10.1186/s12863-023-01123-8", "label": "Paper", "title": "Genomic benchmarks: a collection of datasets for genomic sequence classification" }, "download": { "url": "https://github.com/wyhsleep/GenomicTouchstone", "label": "Benchmark code", "kind": "code" }, "source_url": "https://github.com/wyhsleep/GenomicTouchstone", "publication": { "first_public_date": "2022-06-10", "first_public_url": "https://doi.org/10.1101/2022.06.08.495248", "first_public_platform": "bioRxiv", "venue": "BMC Genomic Data", "venue_year": "2023", "venue_url": "https://doi.org/10.1186/s12863-023-01123-8", "note": "Reference paper for the scratch baseline; not a pretrained-checkpoint release." }, "specifications": { "architecture": "Convolution", "parameters": "—", "input_length": "—", "tokenizer": "Char", "pretraining_objective": "—", "pretraining_data": "—", "input_length_unit": "nt" } }, { "id": "baseline_lstm", "name": "LSTM", "category": "Baselines", "evaluation": "Scratch", "leaderboard_id": "baseline_lstm", "aliases": [ "LSTM" ], "note": "Trained from scratch in this benchmark; no pretrained checkpoint. Paper describes the LSTM architecture.", "paper": { "url": "https://doi.org/10.1162/neco.1997.9.8.1735", "label": "Paper", "title": "Long Short-Term Memory" }, "download": { "url": "https://github.com/wyhsleep/GenomicTouchstone", "label": "Benchmark code", "kind": "code" }, "source_url": "https://github.com/wyhsleep/GenomicTouchstone", "publication": { "first_public_date": "1997-11-01", "first_public_url": "https://doi.org/10.1162/neco.1997.9.8.1735", "first_public_platform": "Neural Computation", "venue": "Neural Computation", "venue_year": "1997", "venue_url": "https://doi.org/10.1162/neco.1997.9.8.1735", "note": "Publication date of the cited journal paper. Reference paper for the scratch baseline; not a pretrained-checkpoint release." }, "specifications": { "architecture": "RNN", "parameters": "—", "input_length": "—", "tokenizer": "Char", "pretraining_objective": "—", "pretraining_data": "—", "input_length_unit": "nt" } }, { "id": "ntv2-50m-random-init", "name": "Basic Transformer", "category": "Baselines", "evaluation": "Scratch", "leaderboard_id": "ntv2-50m-random-init", "aliases": [ "NTv2-50M-Random-Init" ], "note": "Randomly initialized NTv2-50M architecture; shown as NTv2-50M-Random-Init in the leaderboard. No pretrained checkpoint.", "paper": { "url": "https://arxiv.org/abs/1706.03762", "label": "Paper", "title": "Attention is all you need" }, "download": { "url": "https://github.com/wyhsleep/GenomicTouchstone", "label": "Benchmark code", "kind": "code" }, "source_url": "https://github.com/wyhsleep/GenomicTouchstone", "publication": { "first_public_date": "2017-06-12", "first_public_url": "https://arxiv.org/abs/1706.03762v1", "first_public_platform": "arXiv", "venue": "NeurIPS", "venue_year": "2017", "venue_url": "https://arxiv.org/abs/1706.03762", "note": "Reference paper for the scratch baseline; not a pretrained-checkpoint release." }, "specifications": { "architecture": "Attention", "parameters": "55.9M", "input_length": "12,282", "tokenizer": "6-mer", "pretraining_objective": "—", "pretraining_data": "—", "input_length_unit": "nt" } }, { "id": "cnn-56m-random-init", "name": "CNN-56M", "category": "Baselines", "evaluation": "Scratch", "leaderboard_id": "cnn-56m-random-init", "aliases": [ "CNN-56M-Random-Init" ], "note": "Trained from scratch; no pretrained checkpoint. Defined in the revised study Methods; the linked study preprint is an earlier version.", "paper": { "url": "https://doi.org/10.1101/2025.06.25.661622", "label": "Study paper", "title": "Genomic Touchstone: Benchmarking Genomic Language Models in the Context of the Central Dogma" }, "download": { "url": "https://github.com/wyhsleep/GenomicTouchstone", "label": "Benchmark code", "kind": "code" }, "source_url": "https://github.com/wyhsleep/GenomicTouchstone", "publication": { "first_public_date": "2025-06-30", "first_public_url": "https://doi.org/10.1101/2025.06.25.661622", "first_public_platform": "bioRxiv", "venue": "bioRxiv", "venue_year": "2025", "venue_url": "https://doi.org/10.1101/2025.06.25.661622", "note": "Date refers to the linked 2025 study preprint. The CNN-56M baseline is defined in the revised Methods, which postdate that preprint." }, "specifications": { "architecture": "CNN", "parameters": "56.0M", "input_length": "Task-dependent", "tokenizer": "Char", "pretraining_objective": "—", "pretraining_data": "—", "input_length_unit": "nt" } } ] }