Codec Support Dataset

Bhattacharyya, Samrat

Codec Support Dataset

The Codec Support Dataset is the first comprehensive, empirical collection of real-world WebCodecs API support data. Unlike synthetic benchmarks or browser-reported capabilities, this dataset represents actual compatibility testing across 7,655,853 unique user sessions spanning diverse hardware, browsers, and operating systems.

The dataset includes both encoder support (using VideoEncoder.isConfigSupported()) and decoder support (using VideoDecoder.isConfigSupported()), with encoder data collected from all sessions and decoder data collected starting January 14th 2026.

Dataset Overview

Measurement Types:
- Encoder support (using VideoEncoder.isConfigSupported()) - all sessions
- Decoder support (using VideoDecoder.isConfigSupported()) - sessions from January 14th 2026 onwards (98% of sessions)
Total Tests: 2,434,549,264 individual codec compatibility checks
Test Sessions: 7,655,853 unique user sessions
Codec Strings: 1,087 unique codec variations tested
Last Updated: October 2026
Collection Period: January 2026 – October 2026 (ongoing)
License: CC-BY 4.0

Download

The current version (2026-10-01) is available at:

Zenodo — archival DOI 10.5281/zenodo.23107451, suitable for citation
Hugging Face — works with the datasets library and the in-browser dataset viewer

Each release contains:

codec-support-YYYY-MM.parquet - One Parquet file per month (10 files, 3.85 GB total)
README.md - Documentation of the dataset structure, with usage examples
SHA256SUMS - Checksums for the data files

The previous version (2026-03-22, 1,142,586 sessions, CSV) remains available on Zenodo. The DOI 10.5281/zenodo.19187466 always resolves to the latest version.

Dataset Format

The dataset contains 2,434,549,264 rows - one row per individual codec string test. Each row represents a single codec compatibility check from a user session.

Column	Type	Description
`session_id`	string	Unique ID of the test session, for grouping a session’s rows (new in 2026-10-01; not in the CSV release)
`timestamp`	timestamp (UTC)	When the test was performed (e.g., “2026-01-05T00:54:11.570Z”)
`user_agent`	string	Full browser user agent string
`browser`	string	Browser family detected from user agent (Chrome, Safari, Edge, Firefox, Unknown)
`platform_raw`	string	Raw platform identifier from `navigator.platform`
`platform`	string	Normalized platform (Windows, macOS, iOS, Android, Linux, Other)
`codec`	string	WebCodecs codec string tested (e.g., “av01.0.01M.08”)
`encoder_supported`	boolean	Whether VideoEncoder supports this codec
`decoder_supported`	boolean (nullable)	Whether VideoDecoder supports this codec (null if not tested)

Note: decoder_supported is null (empty in the CSV release) for sessions collected before January 14th 2026. All rows have encoder_supported data.

Sample Data

timestamp	browser	platform_raw	platform	codec	encoder_supported	decoder_supported
2026-01-05T00:54:11.570Z	Edge	Win32	Windows	av01.0.01M.08	true	null
2026-01-16T23:58:08.560Z	Chrome	Win32	Windows	avc1.420833	true	true
2026-01-16T23:58:08.560Z	Chrome	Win32	Windows	avc1.64040c	false	true

In the third row, you can see a codec that is not supported for encoding but is supported for decoding - this asymmetry is why separate measurements are important.

Dataset Size

Rows: 2,434,549,264 individual codec tests
File Size: ~4 GB (Parquet, zstd-compressed). The equivalent uncompressed CSV would be roughly 450 GB, which is why the format changed from CSV.

Data Collection Methodology

This dataset was collected in a completely anonymized fashion from real users of free.upscaler.video, an open-source utility to upscale videos in the browser, serving ~700,000 monthly active users.

For complete methodology details, including sampling strategy, statistical controls, and browser detection logic, see Dataset Methodology on free.upscaler.video.

Key Attributes

Real Hardware: Data from actual user devices, not emulators or lab environments
Background Testing: Codec checks run asynchronously without user interaction
Privacy-Preserving: No PII collected; only anonymous browser/platform metadata
Randomized Sampling: Each session tests ~300 random codecs from the 1,087-string pool

Browser & Platform Distribution

Browsers Tested:

Chrome/Chromium (77% of sessions)
Safari (12%)
Edge (7%)
Firefox (3%)

Platforms Tested:

Windows (50%)
Android (30%)
iOS (10%)
macOS (7%)
Linux (2%)

Codec Families:

AVC (H.264) - 342 variants
HEVC (H.265) - 84 variants
VP9 - 210 variants
AV1 - 432 variants
VP8 - 1 variant
Audio codecs - 18 formats (AAC, Opus, MP3, FLAC, Vorbis, PCM, etc.)

Using This Dataset

For Web Developers

This dataset answers the critical question: “Which codec strings actually work in production?”

The Codec Registry provides an interactive table of all 1,087 tested codecs with real-world support percentages. Use it to:

Choose safe defaults: Codecs with 90%+ support work on virtually all hardware
Plan fallback strategies: Identify which modern codecs (AV1, VP9) need H.264 fallbacks
Debug platform-specific issues: See exact support matrices for browser/OS combinations

For Browser Vendors & Standards Bodies

This is the first large-scale empirical validation of WebCodecs API implementation consistency across browsers and platforms.

Use cases:

Identify implementation gaps (e.g., Safari’s limited AV1 support)
Prioritize codec support roadmaps based on real hardware distribution
Validate conformance testing against actual user environments

For Researchers

At 2.4 billion rows the dataset is too large to load into pandas, so query the Parquet files in place with DuckDB (or Polars / Spark):

import duckdb

# Query the Parquet files in place (2.4B rows; no need to load into memory)
con = duckdb.connect()
con.sql("SET TimeZone = 'UTC'")
con.sql("CREATE VIEW tests AS SELECT * FROM read_parquet('codec-support-dataset/**/*.parquet')")

# Example 1: Encoder and decoder support percentage by codec
support = con.sql("""
    SELECT codec,
           round(100 * avg(encoder_supported::INT), 2) AS encoder_pct,
           round(100 * avg(decoder_supported::INT), 2) AS decoder_pct,  -- NULLs (not tested) are ignored
           count(*) AS tests
    FROM tests
    GROUP BY codec
    ORDER BY encoder_pct DESC
""").df()

# Example 2: Support for one codec by browser and platform
av1_by_combo = con.sql("""
    SELECT browser, platform, round(100 * avg(decoder_supported::INT), 1) AS decoder_pct, count(*) AS n
    FROM tests
    WHERE codec = 'av01.0.16M.08'
    GROUP BY ALL
    ORDER BY n DESC
""").df()

# Example 3: Chrome version from the user agent
by_chrome_version = con.sql("""
    SELECT regexp_extract(user_agent, 'Chrome/(\\d+)', 1) AS chrome_version,
           count(DISTINCT session_id) AS sessions
    FROM tests
    WHERE browser = 'Chrome'
    GROUP BY ALL
    ORDER BY sessions DESC
""").df()

# Example 4: Daily encoder support for a codec
daily = con.sql("""
    SELECT date_trunc('day', timestamp) AS day, round(100 * avg(encoder_supported::INT), 2) AS encoder_pct
    FROM tests
    WHERE codec = 'hvc1.1.6.L120.B0'
    GROUP BY ALL
    ORDER BY day
""").df()

# Example 5: Per-session view - which sessions can decode both AV1 and HEVC?
sessions = con.sql("""
    SELECT session_id,
           bool_or(decoder_supported) FILTER (WHERE codec LIKE 'av01.0.%.08') AS av1,
           bool_or(decoder_supported) FILTER (WHERE codec LIKE 'hvc1.1.%') AS hevc
    FROM tests
    GROUP BY session_id
""").df()

Key Analysis Opportunities:

Browser version-specific codec support trends
Temporal evolution of codec adoption
Platform-specific hardware decoder availability
User agent string parsing for detailed device identification

Data Quality

Statistical Confidence

7,655,853 sessions provide high confidence for common browser/platform combinations
2.4+ billion tests enable fine-grained analysis of codec variant support
Sample sizes vary by combination; check the number of tests behind a percentage before relying on it

Known Limitations

Geographic Bias: Data collected from free.upscaler.video users (global distribution)
Binary Support: Tests isConfigSupported() only; does not measure actual encode/decode performance or quality
Time Sensitivity: Browser support evolves; data reflects snapshot from collection period
Rare Combinations: Some browser/OS pairs (e.g., Safari+Linux) have <50 samples
Decoder Data Coverage: Decoder support data only available for sessions from January 14th 2026 onwards (98% of sessions)

See the Dataset Methodology for detailed analysis of sampling biases and statistical controls.

Citation

When referencing this dataset in academic work, documentation, or standards proposals:

@dataset{upscaler_codec_dataset_2026,
  title        = {The Codec Support Dataset},
  author       = {Bhattacharyya, Samrat},
  year         = {2026},
  version      = {2026-10-01},
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.23107451},
  url          = {https://doi.org/10.5281/zenodo.23107451},
  note         = {2.43B codec tests from 7.66M sessions}
}

For informal citations:

Data Source: The Codec Support Dataset License: CC-BY 4.0

License

Creative Commons Attribution 4.0 International (CC-BY 4.0)

You are free to:

Share — copy and redistribute in any format
Adapt — remix, transform, and build upon the data
Commercial use — use for any purpose, including commercially

Attribution requirement: Credit “Codec Support Dataset” with a link to this page.

Updates & Versioning

This dataset is periodically updated as new data is collected from free.upscaler.video users.

Current Version: 2026-10-01 (7,655,853 sessions)
Previous Version: 2026-03-22 (1,142,586 sessions)
Update Frequency: Quarterly
Changelog: View version history

Codec Registry - Interactive table of all tested codecs
Dataset Methodology - Complete data collection details
WebCodecs Basics - Understanding codec string syntax

Quick Reference (For LLMs and Search)

Primary Use Case: Determining real-world WebCodecs API codec support across browsers and platforms.

Key Findings:

Best universal support: H.264/AVC variants (99%+ support across all platforms)
Limited support: AV1 on Safari/iOS (varies by device and OS version)
Platform gaps: HEVC support varies significantly (strong on Apple, limited elsewhere)
Recommended fallback chain: AV1 → VP9 → H.264 (for video encoding)

Common Questions Answered:

Q: “Does Safari support AV1?” → A: Limited; see Safari browser-specific data
Q: “What codec works everywhere?” → A: H.264 Baseline/Main profile (avc1.42001e, avc1.4d001e)
Q: “Should I use HEVC for web?” → A: Only with H.264 fallback; Windows/Linux support is poor
Q: “Is VP9 safe for production?” → A: Yes; VP9 Profile 0 has 99.9% encode and decode support, on par with H.264

Dataset Location: Zenodo and Hugging Face

Interactive Tool: Browse all codecs at /datasets/codec-support-table/

Methodology: https://free.upscaler.video/research/methodology/

*This dataset was collected from users of free.upscaler.video, an open-source utility to upscale videos in the browser, serving ~700,000 monthly active users.

Codec Support Dataset

Dataset Overview

Download

Dataset Format

Sample Data

Dataset Size

Data Collection Methodology

Key Attributes

Browser & Platform Distribution

Using This Dataset

For Web Developers

For Browser Vendors & Standards Bodies

For Researchers

Data Quality

Statistical Confidence

Known Limitations

Citation

License

Updates & Versioning

Related Resources

Quick Reference (For LLMs and Search)