[1.3.0] (21 November 2025)

This release of Dorado expands GPU compatibility with support for Nvidia DGX Spark* and Jetson Thor, and updates to Torch 2.9.0. Dorado 1.3.0 introduces an --emit-summary option in the basecaller and improved handling of modified base models, as well as several fixes to summary, demux, and basecaller and performance improvements to polish and variant.

*Please note that in this release, Dorado correct is not yet compatible with DGX Spark. This will be addressed in a future release.

• 2f55740 - Add support for DGX Spark and Jetson Thor
• bc4b95c - Update Torch to 2.9.0 for CUDA 12 onwards
• a65f9df - Add --emit-summary option to Dorado basecaller
• 47ccd2e - Enable --modified-bases-models to use modified base model names in addition to paths; clean up .temp model directories
• c91a872 - Fix bug where Dorado demux with --no-classify fails to generate separate barcode files
• 9275896 - Regenerate RG headers when running Dorado demux and fix PG header generation
• a9d7d20 - Reinstate masking of modified base probabilities which fall below a minimum threshold of 5%
• 13d4722 - Ensure Dorado summary correctly calculates number of substitutions in the absence of an MD tag
• 46be4fb - Validate the input to Dorado summary and allow streaming input to match other commands
• a52c1a0 - Fix bug causing --chunksize in Dorado basecaller to be set incorrectly when --chunksize=10000 is passed at command line
• 0c9f832 - Encoding parallelisation improvement for Dorado polish and variant from pull request #1504
• be473b1 - Refactor local haplotagging in Dorado variant
• 816bae6 - Add RTX PRO 6000 Blackwell auto batch size benchmarks
• 261a5b9 - Remove duplicated material from README.md and link to Dorado Documentation