[2.0.0] (20 May 2026)

This major release of Dorado includes new v6.0 basecalling models for DNA and RNA, state-of-the-art small variant calling capabilities with Dorado smallvar (formerly Dorado variant), and a range of other performance, usability, and stability improvements. This release introduces:

• New v6.0 basecalling models
  • A more accurate basecalling model (HAC v6.0) for DNA
    • High single-molecule accuracy with improvements in short tandem repeat expansions, homopolymers, and low-complexity regions
    • Improved accuracy in bacterial methylation motifs, enabling polished assemblies comparable to SUP v5.2
    • Updated modified base models for 5mC/5hmC, 4mC/5mC, and 6mA, with substantial improvements in modified base accuracy
  • Updated fast, HAC, and SUP models for RNA basecalling
    • Improved headline single-molecule accuracy, particularly for HAC basecalling
    • Updated models for RNA modifications, including substantial gains for HAC pseU, HAC m6A/inosine, and SUP m5C/2OmeC
  • Updated models for Dorado polish
    • New polishing model for v6.0 HAC added
    • Bacterial polishing model now also supports v6.0 HAC input data
• New Dorado smallvar tool replacing Dorado variant for SNP and INDEL calling
  • Dorado smallvar provides state-of-the-art variant calling accuracy from data basecalled with DNA HAC v6.0 and HAC v5.2 models
  • Dorado smallvar is significantly faster than Dorado variant through a novel model architecture, highly optimised GPU execution, and scalability across multiple GPU devices

Other fixes and improvements have been implemented in Dorado aligner, demux, summary, and more, with the full list of changes below.

• eba46ab - Add v6.0 DNA and RNA basecalling, modified base, and Dorado polish models, including new HAC basecall model architecture
• 0d519ea - Replace Dorado variant with Dorado smallvar
• 23e04b8 - Dorado smallvar support for v6.0 and v5.2 HAC models
• aa90f44 - Replace --model with --models-directory and --model-override in Dorado smallvar
• cb41979 - Add a --model-override option to Dorado polish
• 0f5e090 - Dorado polish now processes only the draft sequences from the input BAM file header
• f309055 - Enable the -eqx minimap2 option in Dorado aligner
• ad3f074 - Prevent excessive memory usage when running alignment
• eb1efa8 - Prevent over-splitting of RNA reads
• 49e9f07 - Improve speed of Dorado summary
• 2ed0f11 - Fix summary file output to better match results from Dorado basecall server and SAM specification
• 0fce0e0 - Initial support for custom barcoding arrangements with dual barcodes
• ae2a8b6 - Improve handling of inconsistent read group data in Dorado demux and aligner
• 0edb926 - Remove support for consuming stdin for Dorado demux
• 23f944a - Fix Dorado demux crash when running on previously demultiplexed files
• ef3835f - Fix Dorado demux crash when processing previously unclassified read groups with --no-trim flag
• d68d51d - Include tm:Z tag in output file RG lines to indicate trim conditions
• b47e64d - Prevent Dorado basecaller crash when no sample_id in POD5 file
• 757b250 - Fix output folder paths to include experiment ID when processing POD5s that were converted from .fast5
• f3758e1 - Increase minimum supported CUDA version to 12.8 (or 12.6 on Linux ARM)
• 4a4ca7a - Update to OpenSSL 3.5.5
• 9bb7296 - Update to pod5 0.3.39
• 4e39b86 - Dorado correct now defaults to CPU mode on macOS but allows a user-specified --device metal
• ceb4ba8 - Add pre-computed batch size benchmarks for RTX PRO 6000 Workstation cards