OPUS vs The Stack v2

Provides a web-based search interface that queries a database index across 1,214 distinct parallel corpora spanning 1,005 languages, allowing users to filter by language pair and corpus type to identify relevant training data. The discovery system aggregates metadata (sentence pair counts, corpus source, release dates) from heterogeneous sources including subtitles, institutional documents, and web crawls, presenting results ranked by corpus size and relevance.

Unique: Aggregates and indexes 1,214 distinct corpora from heterogeneous sources (subtitles, EU documents, web crawls, academic sources) into a unified searchable interface, rather than requiring users to visit individual corpus repositories. Maintains version tracking across releases (e.g., OpenSubtitles v2024 vs historical versions) and exposes corpus composition percentages relative to the full 102.9B sentence pair collection.

vs alternatives: Broader corpus coverage (1,214 corpora, 1,005 languages) than single-source alternatives like OpenSubtitles alone, but lacks the quality filtering, alignment confidence scores, and API-based programmatic access that commercial MT platforms provide.

Enables download of aligned sentence pairs from selected corpora in their native format, aggregating data from 102.9 billion total sentence pairs across sources like OpenSubtitles (27.2B), NLLB (22.7B), CCMatrix (17.1B), and 1,209 additional corpora. Downloads are organized hierarchically by corpus and language pair, with file formats and encoding specifications determined by the source corpus (format specifications not explicitly documented in available materials).

Unique: Aggregates downloads from 1,214 distinct corpora with heterogeneous sources and formats into a unified interface, allowing single-point access to subtitle data (OpenSubtitles 27.2B pairs), institutional documents (EU Europarl 217.4M, DGT 1.2B), web-crawled data (CCMatrix 17.1B, ParaCrawl 4.6B), and domain-specific corpora (medical EMEA 282.5M, patents EuroPat 252.2M). Maintains version history with release tracking (e.g., OpenSubtitles v2024 released 2025-02-14).

vs alternatives: Provides access to 102.9B sentence pairs across 1,005 languages in a single interface, whereas alternatives like individual corpus repositories require visiting multiple sites; however, lacks programmatic API access, quality filtering, and explicit licensing documentation that commercial MT data providers offer.

Provides access to specialized domain-specific parallel corpora including EMEA (medical, 282.5M pairs), EuroPat (patents, 252.2M), and Bible translations (88.3M), enabling training of translation systems for specialized domains with domain-specific terminology and language patterns. These corpora are sourced from authoritative domain-specific documents and enable building translation systems for vertical markets.

Unique: Aggregates specialized domain-specific corpora including EMEA (medical, 282.5M pairs), EuroPat (patents, 252.2M), and Bible translations (88.3M), providing domain-specific parallel data for vertical markets. While small relative to general-domain corpora, these specialized sources enable training of domain-specific translation systems with domain-specific terminology and language patterns.

vs alternatives: Provides centralized access to specialized domain corpora in a single interface, whereas accessing these sources individually requires visiting domain-specific repositories; however, limited domain coverage (only medical, patents, Bible) and small corpus sizes mean specialized MT platforms with broader domain coverage and larger domain-specific datasets are more suitable for most vertical markets.

Enables users to identify and download parallel corpora organized by domain and source type, including subtitle-based data (OpenSubtitles, TED talks), institutional/legal documents (EU Europarl, JRC-Acquis, DGT), web-crawled general-domain data (CCMatrix, ParaCrawl, WikiMatrix), and specialized corpora (medical EMEA, patents EuroPat, Bible translations). The collection exposes corpus composition metadata allowing users to understand source characteristics and select data matching their domain requirements.

Unique: Curates domain-specific corpora including medical (EMEA 282.5M pairs), patents (EuroPat 252.2M), legal/institutional (Europarl 217.4M, JRC-Acquis 215.9M, DGT 1.2B), and specialized sources (Bible translations 88.3M, Ubuntu documentation) alongside general-domain subtitle and web-crawled data, enabling users to select data by source type and implied domain rather than explicit domain labels.

vs alternatives: Provides access to specialized domain corpora (medical, legal, patents) in a single interface, whereas generic parallel corpus repositories focus on general-domain data; however, lacks explicit domain tagging, quality metrics per domain, and domain-specific preprocessing that specialized MT data providers offer.

Exposes corpus-level metadata including total sentence pair counts, percentage of collection, source type, and release dates, enabling users to understand the composition and scale of available parallel data. Provides aggregate statistics showing that top 10 corpora account for ~93.5% of total data, with detailed breakdowns for major sources (OpenSubtitles 27.2B/26.47%, NLLB 22.7B/22.09%, CCMatrix 17.1B/16.61%, ParaCrawl 4.6B/4.50%).

Unique: Aggregates and exposes composition statistics across 1,214 corpora totaling 102.9B sentence pairs, showing that top 10 corpora represent ~93.5% of data and identifying the long tail of 1,200+ corpora with minimal coverage. Provides per-corpus metadata (sentence pair counts, percentages, release dates) enabling data-driven selection, rather than requiring users to assess corpus sizes individually.

vs alternatives: Offers transparent composition statistics across a large aggregated collection, whereas individual corpus repositories provide only their own metrics; however, lacks per-language-pair breakdowns, quality-weighted statistics, and temporal trend analysis that research-focused data platforms provide.

Maintains version history for major corpora with explicit release dates, enabling users to access specific versions for reproducibility and comparative analysis. Tracks releases including OpenSubtitles v2024 (released 2025-02-14), HPLT and MultiHPLT v2 (released 2025-01-25), and historical versions back to 2017, allowing researchers to reproduce results with the same data version used in prior work.

Unique: Explicitly tracks and maintains version history for major corpora with release dates (e.g., OpenSubtitles v2024 released 2025-02-14, HPLT v2 released 2025-01-25), enabling reproducible research and comparative analysis across versions. Provides historical access to corpus versions dating back to 2017, rather than only offering the latest version.

vs alternatives: Enables version-based reproducibility for major corpora, whereas many corpus repositories only provide the latest version; however, lacks detailed changelogs, automated version management, and integration with ML experiment tracking tools that research platforms like Hugging Face Datasets provide.

Aggregates parallel data for 1,005 languages including low-resource and endangered languages, though with highly uneven coverage. Provides access to specialized multilingual corpora (MultiHPLT 2.7B pairs, MultiParaCrawl 2.8B, MultiCCAligned 2.4B) designed to cover broader language sets, alongside language-specific corpora for rare pairs. However, the long tail of 1,200+ corpora with minimal coverage means many language pairs have severely limited data.

Unique: Aggregates data for 1,005 languages including low-resource and endangered languages, with specialized multilingual corpora (MultiHPLT 2.7B, MultiParaCrawl 2.8B, MultiCCAligned 2.4B) designed to provide broader language coverage. However, coverage is highly uneven with top 3 corpora representing 65.17% of data, meaning most rare language pairs have minimal or zero coverage.

vs alternatives: Provides access to 1,005 languages in a single interface, whereas most MT platforms focus on high-resource pairs; however, the uneven distribution and lack of explicit language pair availability matrix make it difficult to assess coverage for specific rare pairs, and data quality for low-resource languages is undocumented.

Provides access to large-scale institutional and legal parallel corpora sourced from EU documents and similar official sources, including Europarl (217.4M pairs), JRC-Acquis (215.9M), DGT (1.2B), and similar sources. These corpora contain formal, high-quality aligned sentence pairs from official multilingual documents, suitable for training translation systems on institutional and legal language.

Unique: Aggregates large-scale institutional and legal parallel corpora from EU sources (Europarl 217.4M, JRC-Acquis 215.9M, DGT 1.2B) providing high-quality formal language data from official multilingual documents. DGT corpus alone (1.2B pairs) represents 1.17% of total OPUS collection, making institutional data a significant component of the aggregation.

vs alternatives: Provides centralized access to EU institutional corpora in a single interface, whereas accessing these sources individually requires navigating multiple government and institutional repositories; however, lacks domain-specific filtering, quality metrics, and documentation of preprocessing applied to institutional documents.

Aggregates 67 TB of source code from the Software Heritage archive, filtering for permissively licensed repositories (MIT, Apache 2.0, BSD, etc.) across 600+ programming languages. Uses automated license detection and validation to ensure legal compliance for model training. Implements a rigorous deduplication pipeline at file and repository levels to eliminate redundant training data and reduce dataset bloat.

Unique: Largest open-source code dataset at 67 TB with automated opt-out governance allowing repository owners to request removal, combined with rigorous deduplication and PII removal pipeline — no other public dataset offers this scale with legal compliance and community control mechanisms

vs alternatives: Larger and more legally compliant than GitHub's CodeSearchNet (14M files) or Google's BigQuery public datasets, with explicit opt-out governance vs. implicit inclusion, and covers 600+ languages vs. Codex training data's undisclosed language distribution

Implements a community-driven opt-out system where repository owners can request removal of their code from the dataset without legal takedown notices. Maintains a registry of excluded repositories and re-applies exclusions during dataset updates. Provides transparent governance documentation and a clear submission process for removal requests, balancing open access with creator rights.

Unique: First large-scale code dataset to implement opt-out governance at dataset level rather than relying solely on license compliance, with transparent registry and community submission process — shifts power from dataset creators to code contributors

vs alternatives: More respectful of creator autonomy than GitHub Copilot's training approach (no opt-out) or academic datasets (one-time snapshot), and more scalable than individual DMCA takedowns

Automated pipeline that scans source code for personally identifiable information (email addresses, API keys, SSH keys, credit card patterns, phone numbers) and removes or redacts them before dataset release. Uses regex patterns, entropy-based detection for secrets, and heuristic rules to identify sensitive data. Operates at file level with configurable sensitivity thresholds to balance data utility against privacy risk.

Unique: Combines regex pattern matching, entropy-based secret detection, and heuristic rules in a unified pipeline with configurable sensitivity — more comprehensive than simple regex-only approaches, but trades off false positive rate against security coverage

vs alternatives: More thorough than GitHub's secret scanning (which only flags known patterns) because it includes entropy-based detection for unknown secret formats, but less accurate than specialized tools like TruffleHog due to language-agnostic approach

Indexes 67 TB of source code across 600+ programming languages with language-aware metadata (syntax, file extension, language family). Enables retrieval by language, license, repository, or code patterns. Uses Software Heritage's existing indexing infrastructure as foundation, augmented with language detection and classification. Supports both bulk download and filtered queries for specific language subsets.

Unique: Leverages Software Heritage's existing language detection and indexing infrastructure, then augments with BigCode-specific language classification and filtering — avoids reinventing language detection while providing dataset-specific query capabilities

vs alternatives: More comprehensive language coverage (600+ languages) than GitHub's Linguist (500+ languages) and more accessible than Software Heritage's raw API because it's pre-filtered for permissive licenses and deduplicated

Removes duplicate code files and repositories using content hashing (SHA-256 or similar) and fuzzy matching for near-duplicates. Operates in two stages: exact deduplication via hash matching, then fuzzy matching (e.g., Jaccard similarity or MinHash) to catch semantically identical code with minor formatting differences. Preserves one canonical copy of each unique code pattern while removing redundant training examples.

Unique: Two-stage deduplication combining exact hash matching with fuzzy similarity matching (likely MinHash or Jaccard) to catch both identical and near-identical code — more thorough than single-stage approaches but computationally expensive

vs alternatives: More aggressive deduplication than CodeSearchNet (which uses simple hash matching) because it catches near-duplicates, but less semantic than clone detection tools (which understand code structure) because it's content-based

Integrates with Software Heritage's comprehensive archive of 200+ million repositories and their full version control history. Extracts source code snapshots from Software Heritage's Git/Mercurial/SVN repositories, preserving repository metadata (commit history, author info, timestamps). Provides access to code at specific points in time, enabling historical analysis or training on code evolution patterns.

Unique: Leverages Software Heritage's universal code archive (200M+ repositories) as data source, providing access to code that would be impossible to collect via GitHub API alone — enables training on archived/deleted repositories and non-GitHub platforms (GitLab, Gitea, etc.)

vs alternatives: More comprehensive than GitHub-only datasets because it includes code from GitLab, Gitea, SourceForge, and other platforms archived by Software Heritage; more legally defensible than web scraping because it uses an established, community-maintained archive

Tracks and validates SPDX license identifiers for each repository, ensuring only permissively licensed code (MIT, Apache 2.0, BSD, etc.) is included. Maintains license metadata alongside code files, enabling downstream users to verify legal compliance. Implements license hierarchy and compatibility checking to handle dual-licensed or complex licensing scenarios.

Unique: Combines automated SPDX detection with manual review and maintains license metadata alongside code, enabling downstream users to verify compliance — more transparent than datasets that simply claim 'permissive licenses' without proof

vs alternatives: More legally rigorous than GitHub's CodeSearchNet (which doesn't validate licenses) and more transparent than Codex training data (which doesn't disclose license filtering at all)

Maintains versioned snapshots of the dataset (e.g., v2.0, v2.1) with documented changes between versions (new repositories added, deduplication improvements, PII removal updates). Provides checksums and manifests for reproducibility, enabling researchers to cite specific dataset versions and reproduce results. Tracks dataset lineage and transformation history.

Unique: Maintains semantic versioning and detailed changelogs for dataset releases, enabling researchers to cite specific versions and understand dataset evolution — more rigorous than one-off dataset releases without versioning

vs alternatives: More reproducible than academic datasets that are released once without versioning, and more transparent than commercial datasets (Codex) that don't disclose version history or changes