Why DuckDB 2.0 is faster
Points and comments are a snapshot, not live.
DuckDB 2.0 claims 2-3x faster S3 reads via async I/O and improved recursive CTEs.
DuckDB 2.0 alpha shows 2-3x speedup reading Parquet/CSV from S3 by separating download and decode threads. Recursive CTE engine rewritten: walking a 20,000-commit git ancestry dropped from ~16s to 0.10s. New VARIANT type shreds consistent JSON fields into real columns, achieving 2.7x storage reduction and 6x faster field queries vs JSON strings. Also added triggers, nested schemas, DML in CTEs, Spark SQL compatibility mode, spill-to-disk remote file cache, and CLI SQL formatter.
What commenters are saying
Multiple commenters flagged heavy LLM tells in the article's prose ("The cost is now about the rows you actually touch", "does not rescue it"), finding it hard to read and suspecting the claims were not thoroughly vetted. One commenter noted the 10,000-commit gap since 1.5 seemed high for human work. A few defended the content as valuable despite style concerns. Others suggested the writing may be by an ESL speaker mixed with AI drafting.