Model type
Autoregressive DNA model with StripedHyena 2
Configuration
Evo2
Evo 2 models and generates DNA over long contexts at single-nucleotide resolution.
Sources (4)
ArcInstitute/evo2: README.md; ArcInstitute/evo2_7b: README.md; ArcInstitute/evo2_7b: config.json; evo2: Journal full-text XML · Evo 2 paper: architecture, training, and data, Figure 1 and Discussion; official checkpoints table and evo2_7b licence metadataHow it worksEvo 2 workflow
Conceptual summary of the documented data flow; optional inputs and configured downstream stages must be reported for a reproducible evaluation.
Sources (4)
ArcInstitute/evo2: README.md; ArcInstitute/evo2_7b: README.md; ArcInstitute/evo2_7b: config.json; evo2: Journal full-text XML · Evo 2 paper: architecture, training, and data, Figure 1 and Discussion; official checkpoints table and evo2_7b licence metadataOverview
Inputs
DNA sequences represented at single-base resolution.
Outputs
Next-token outputs, embeddings and generated DNA sequences.
Access
Official project documentation and implementation: https://github.com/ArcInstitute/evo2
Sources (4)
ArcInstitute/evo2: README.md; ArcInstitute/evo2_7b: README.md; ArcInstitute/evo2_7b: config.json; evo2: Journal full-text XML · Evo 2 paper: architecture, training, and data, Figure 1 and Discussion; official checkpoints table and evo2_7b licence metadatalimited source coverage · Automated source review, 2026-09-16. All specifications and missing details
Evaluations and results
10 evaluations · 10 results. Different protocols are not a single leaderboard.
Filter evaluations
Applied filters: All linked evaluations
| Tested configuration | Protocol and dataset | Finding | Evidence and details |
|---|---|---|---|
| Configuration: Evo2 | Task: mRNABench ECLIP: eCLIP binding site prediction Dataset subset: mRNABench eCLIP (mRNABench split) | 47.3% auprc percent · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo2 on mRNABench ECLIP: eCLIP binding site prediction A linear probe over frozen embeddings, scored as the mean over ten random seeds. Aggregation: Not reported mRNABench: A curated benchmark for mature mRNA property and function prediction · Table 2, row(Evo2), column(eCLIP) |
| Configuration: Evo2 | Task: mRNABench GO: Gene Ontology term prediction Dataset subset: mRNABench GO (mRNABench split) | 46.7% auprc percent · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo2 on mRNABench GO: Gene Ontology term prediction A linear probe over frozen embeddings, scored as the mean over ten random seeds. Aggregation: Not reported mRNABench: A curated benchmark for mature mRNA property and function prediction · Table 2, row(Evo2), column(GO) |
| Configuration: Evo2 | Task: mRNABench HL: mRNA half life Dataset subset: mRNABench HL (mRNABench split) | 0.66 pearson_r correlation · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo2 on mRNABench HL: mRNA half life A linear probe over frozen embeddings, scored as the mean over ten random seeds. Aggregation: Not reported mRNABench: A curated benchmark for mature mRNA property and function prediction · Table 2, row(Evo2), column(HL) |
| Configuration: Evo2 | Task: mRNABench MRL-HL-PAIR: Paired mean ribosome load and half life Dataset subset: mRNABench MRL-HL-Pair (mRNABench split) | 0.51 pearson_r correlation · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo2 on mRNABench MRL-HL-PAIR: Paired mean ribosome load and half life A linear probe over frozen embeddings, scored as the mean over ten random seeds. Aggregation: Not reported mRNABench: A curated benchmark for mature mRNA property and function prediction · Table 2, row(Evo2), column(MRL-HL-Pair) |
| Configuration: Evo2 | Task: mRNABench MRL-MPRA: Mean ribosome load on an MPRA library Dataset subset: mRNABench MRL MPRA (mRNABench split) | 0.7 pearson_r correlation · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo2 on mRNABench MRL-MPRA: Mean ribosome load on an MPRA library A linear probe over frozen embeddings, scored as the mean over ten random seeds. Aggregation: Not reported mRNABench: A curated benchmark for mature mRNA property and function prediction · Table 2, row(Evo2), column(MRL MPRA) |
| Configuration: Evo2 | Task: mRNABench MRL: Mean ribosome load Dataset subset: mRNABench MRL (mRNABench split) | 0.45 pearson_r correlation · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo2 on mRNABench MRL: Mean ribosome load A linear probe over frozen embeddings, scored as the mean over ten random seeds. Aggregation: Not reported mRNABench: A curated benchmark for mature mRNA property and function prediction · Table 2, row(Evo2), column(MRL) |
| Configuration: Evo2 | Task: mRNABench MRNA-LOC-LR: mRNA localisation, long range Dataset subset: mRNABench mRNA Loc-LR (mRNABench split) | 79.5% auprc percent · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo2 on mRNABench MRNA-LOC-LR: mRNA localisation, long range A linear probe over frozen embeddings, scored as the mean over ten random seeds. Aggregation: Not reported mRNABench: A curated benchmark for mature mRNA property and function prediction · Table 2, row(Evo2), column(mRNA Loc-LR) |
| Configuration: Evo2 | Task: mRNABench MRNA-LOC-SR: mRNA localisation, short range Dataset subset: mRNABench mRNA Loc-SR (mRNABench split) | 76.7% auprc percent · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo2 on mRNABench MRNA-LOC-SR: mRNA localisation, short range A linear probe over frozen embeddings, scored as the mean over ten random seeds. Aggregation: Not reported mRNABench: A curated benchmark for mature mRNA property and function prediction · Table 2, row(Evo2), column(mRNA Loc-SR) |
| Configuration: Evo2 | Task: mRNABench PROT-LOC: Protein localisation Dataset subset: mRNABench Prot Loc (mRNABench split) | 40.8% auprc percent · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo2 on mRNABench PROT-LOC: Protein localisation A linear probe over frozen embeddings, scored as the mean over ten random seeds. Aggregation: Not reported mRNABench: A curated benchmark for mature mRNA property and function prediction · Table 2, row(Evo2), column(Prot Loc) |
| Configuration: Evo2 | Task: mRNABench VEP: Variant effect prediction Dataset subset: mRNABench VEP (mRNABench split) | 32.1% auprc percent · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo2 on mRNABench VEP: Variant effect prediction A linear probe over frozen embeddings, scored as the mean over ten random seeds. Aggregation: Not reported mRNABench: A curated benchmark for mature mRNA property and function prediction · Table 2, row(Evo2), column(VEP) |
Source checking is not independent reproduction. Release 2026-09-29-06401fd5b220.
Use this model
How it works, versions and access
Related profile: Evo 2. This page retains the exact record and its evaluation context.
This configuration
Best checkpoint of this model family, as selected by the mRNABench authors and listed in their Appendix D.
- record
- Evo2
- configuration
- Not reported
- entity type
- Configuration
How it works
How it works
Evo 2 models and generates DNA over long contexts at single-nucleotide resolution. StripedHyena 2 hybrid architecture combining short, medium and long convolution operators with attention, trained autoregressively at single-base resolution. The documented inputs are DNA sequences represented at single-base resolution. The output consists of next-token outputs, embeddings and generated DNA sequences.
Sources (4)
ArcInstitute/evo2: README.md; ArcInstitute/evo2_7b: README.md; ArcInstitute/evo2_7b: config.json; evo2: Journal full-text XML · Evo 2 paper: architecture, training, and data, Figure 1 and Discussion; official checkpoints table and evo2_7b licence metadataVersions and reproducibility
Base 8K models, long-context 1M models, 7B 262K model and separately fine-tuned Microviridae model. Checkpoint-dependent: 8K, 262K or 1M bases as listed in the Checkpoints table.
Sources (4)
ArcInstitute/evo2: README.md; ArcInstitute/evo2_7b: README.md; ArcInstitute/evo2_7b: config.json; evo2: Journal full-text XML · Evo 2 paper: architecture, training, and data, Figure 1 and Discussion; official checkpoints table and evo2_7b licence metadataStrengths, limitations and unresolved questions
Strengths and limitations
Strengths and considerations
- Different released context lengths and model scales support a range of sequence modeling workflows.
Sources (4)
ArcInstitute/evo2: README.md; ArcInstitute/evo2_7b: README.md; ArcInstitute/evo2_7b: config.json; evo2: Journal full-text XML · Evo 2 paper: architecture, training, and data, Figure 1 and Discussion; official checkpoints table and evo2_7b licence metadata
Limitations and conditions
- Hardware requirements differ by checkpoint: the README requires FP8/Transformer Engine and Hopper GPUs for some scales, while 7B supports bfloat 16 on a wider set of GPUs.
Sources (4)
ArcInstitute/evo2: README.md; ArcInstitute/evo2_7b: README.md; ArcInstitute/evo2_7b: config.json; evo2: Journal full-text XML · Evo 2 paper: architecture, training, and data, Figure 1 and Discussion; official checkpoints table and evo2_7b licence metadata
Profile review details
Inspected pinned official documentation, relevant implementation files and named primary-paper sections. Claims are limited to those artifacts. Remaining field extraction and identity conflicts are explicit; no new performance claims, model runs or human review are implied.
Stable record: discovery-model-evo-2Specifications
Inputs, training, access and other details
Explanatory profile: limited source coverage · Automated source review, 2026-09-16. Review applies to the cited claims; unresolved fields are listed below. Numerical results retain their own review status.
| Property | Description and evidence |
|---|---|
| Model type | Autoregressive DNA model with StripedHyena 2Sources (4)ArcInstitute/evo2: README.md; ArcInstitute/evo2_7b: README.md; ArcInstitute/evo2_7b: config.json; evo2: Journal full-text XML · Evo 2 paper: architecture, training, and data, Figure 1 and Discussion; official checkpoints table and evo2_7b licence metadata |
| Architecture | StripedHyena 2 hybrid architecture combining short, medium and long convolution operators with attention, trained autoregressively at single-base resolution.Sources (4)ArcInstitute/evo2: README.md; ArcInstitute/evo2_7b: README.md; ArcInstitute/evo2_7b: config.json; evo2: Journal full-text XML · Evo 2 paper: architecture, training, and data, Figure 1 and Discussion; official checkpoints table and evo2_7b licence metadata |
| Inputs | DNA sequences represented at single-base resolution.Sources (4)ArcInstitute/evo2: README.md; ArcInstitute/evo2_7b: README.md; ArcInstitute/evo2_7b: config.json; evo2: Journal full-text XML · Evo 2 paper: architecture, training, and data, Figure 1 and Discussion; official checkpoints table and evo2_7b licence metadata |
| Outputs | Next-token outputs, embeddings and generated DNA sequences.Sources (4)ArcInstitute/evo2: README.md; ArcInstitute/evo2_7b: README.md; ArcInstitute/evo2_7b: config.json; evo2: Journal full-text XML · Evo 2 paper: architecture, training, and data, Figure 1 and Discussion; official checkpoints table and evo2_7b licence metadata |
| Parameters | 1B, 7B, 20B and 40B checkpoints are listed.Sources (4)ArcInstitute/evo2: README.md; ArcInstitute/evo2_7b: README.md; ArcInstitute/evo2_7b: config.json; evo2: Journal full-text XML · Evo 2 paper: architecture, training, and data, Figure 1 and Discussion; official checkpoints table and evo2_7b licence metadata |
| Known versions | Base 8K models, long-context 1M models, 7B 262K model and separately fine-tuned Microviridae model.Sources (4)ArcInstitute/evo2: README.md; ArcInstitute/evo2_7b: README.md; ArcInstitute/evo2_7b: config.json; evo2: Journal full-text XML · Evo 2 paper: architecture, training, and data, Figure 1 and Discussion; official checkpoints table and evo2_7b licence metadata |
| Training data | OpenGenome2 contains more than 8.8T curated nucleotides across bacteria, archaea, eukaryotes and bacteriophage. The paper separates 2.4T tokens of training exposure for 7B from 9.3T for 40B; eukaryotic-host viral sequences were excluded.Sources (4)ArcInstitute/evo2: README.md; ArcInstitute/evo2_7b: README.md; ArcInstitute/evo2_7b: config.json; evo2: Journal full-text XML · Evo 2 paper: architecture, training, and data, Figure 1 and Discussion; official checkpoints table and evo2_7b licence metadata |
| Training cutoff | OpenGenome2 combines multiple nucleotide collections. The inspected paper and released checkpoint documentation do not provide one latest-deposition date that covers every component. · Not reported in inspected sourcesSources (4)ArcInstitute/evo2: README.md; ArcInstitute/evo2_7b: README.md; ArcInstitute/evo2_7b: config.json; evo2: Journal full-text XML · Evo 2 paper: architecture, training, and data, Figure 1 and Discussion; official checkpoints table and evo2_7b licence metadata |
| Context limits | Checkpoint-dependent: 8K, 262K or 1M bases as listed in the Checkpoints table.Sources (4)ArcInstitute/evo2: README.md; ArcInstitute/evo2_7b: README.md; ArcInstitute/evo2_7b: config.json; evo2: Journal full-text XML · Evo 2 paper: architecture, training, and data, Figure 1 and Discussion; official checkpoints table and evo2_7b licence metadata |
| Weights licence | Apache-2.0 is declared in the inspected ArcInstitute/evo2_7b model card; other checkpoints require their own pinned terms.Sources (4)ArcInstitute/evo2: README.md; ArcInstitute/evo2_7b: README.md; ArcInstitute/evo2_7b: config.json; evo2: Journal full-text XML · Evo 2 paper: architecture, training, and data, Figure 1 and Discussion; official checkpoints table and evo2_7b licence metadata |
| Access | Official project documentation and implementation: https://github.com/ArcInstitute/evo2Sources (4)ArcInstitute/evo2: README.md; ArcInstitute/evo2_7b: README.md; ArcInstitute/evo2_7b: config.json; evo2: Journal full-text XML · Evo 2 paper: architecture, training, and data, Figure 1 and Discussion; official checkpoints table and evo2_7b licence metadata |
| Code licence | Apache-2.0SourcesArcInstitute/evo2: LICENSE · LICENSE: licence text |
Evidence
Source checking verifies the cited claim or transcription. It does not establish independent reproduction.
Evidence table
Inspect claims, sources and review details
Trace each statement to its source and review. A context-only reference supports the record generally; it does not verify an individual field. Source checking does not reproduce an experiment.
One row per statement and cited source. Multiple citations are not independent evaluations. Shared locators are labelled explicitly.
2 evidence rows matching the loaded filters
| Property and statement | Original source and location | Review and provenance |
|---|---|---|
| Relationship: family discovery-model-evo-2 Individual claims | ArcInstitute/evo2: README.md Primary paper model-inventory and results tables: mRNABench Table 2; NABench model inventory; LAMBDA Table 1 (EVO2 7B); source-labelled configuration Evo2 | Existing reviewed locator: Table 2, row(Evo2) Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: 53f195997257c56c00e5ef8d33a54f5baad143a6 | source checked automated source review · 2026-09-23 Audit detailsSource review establishes this relationship only. Exact evaluated configurations and original numerical review status remain unchanged. Explicit Evo2/EVO2 identity supported by cited paper and official Evo2 checkpoint list; base and context-extended configurations remain distinct. Field: Claim: model-evaluation-identity-f5f9a62b6ed58e0b4210 Source artifact SHA-256: Hash scope: SHA-256 of retrieved original artifact bytes Format: original_artifact |
| Relationship: family discovery-model-evo-2 Individual claims | mRNABench: A curated benchmark for mature mRNA property and function prediction Primary paper model-inventory and results tables: mRNABench Table 2; NABench model inventory; LAMBDA Table 1 (EVO2 7B); source-labelled configuration Evo2 | Existing reviewed locator: Table 2, row(Evo2) Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: preprint archived 2025-07-08 | source checked automated source review · 2026-09-23 Audit detailsSource review establishes this relationship only. Exact evaluated configurations and original numerical review status remain unchanged. Explicit Evo2/EVO2 identity supported by cited paper and official Evo2 checkpoint list; base and context-extended configurations remain distinct. Field: Claim: model-evaluation-identity-f5f9a62b6ed58e0b4210 Source artifact SHA-256: Hash scope: Exact retrieved primary paper artifact bytes. |
Sources and history
View linked audit checks and correction history
Release 2026-09-29-06401fd5b220 · Record review: source checked
1 source records and release history
- mRNABench: A curated benchmark for mature mRNA property and function prediction · Original source · preprint archived 2025-07-08
Technical metadata and extraction receipts
Stable ID: mrnabench-method-evo2
- areas
- rna-transcriptomes
- source locator
- Table 2, row(Evo2)
- missing metadata
- checkpoint revision: unreported; parameters: unextracted
Related records
- family: Evo 2
- subject: Evo2: family Evo 2
- model: Evo2 on mRNABench ECLIP: eCLIP binding site prediction
- model: Evo2 on mRNABench GO: Gene Ontology term prediction
- model: Evo2 on mRNABench HL: mRNA half life
- model: Evo2 on mRNABench MRL: Mean ribosome load
- model: Evo2 on mRNABench MRL-HL-PAIR: Paired mean ribosome load and half life
- model: Evo2 on mRNABench MRL-MPRA: Mean ribosome load on an MPRA library
- model: Evo2 on mRNABench MRNA-LOC-LR: mRNA localisation, long range
- model: Evo2 on mRNABench MRNA-LOC-SR: mRNA localisation, short range
- model: Evo2 on mRNABench PROT-LOC: Protein localisation
- model: Evo2 on mRNABench VEP: Variant effect prediction