ClickHouse v26.7 Release Webinar

The ClickHouse v26.7 release webinar covers a genuinely massive summer release — 64 new features, 117 performance optimizations (the first release ever to crack 100), and over 300 bug fixes. Alexey Milovidov races through the highlights, with live demos of the open-source SQL console, in-terminal autocompletion, built-in offline docs, and — as a bonus — a full ray tracer written entirely in ClickHouse SQL rendering 4K images and video. 🔸 Code in any language — run Python, C, or Rust snippets inside SQL via configurable drivers, sandboxed with gVisor so untrusted code (even code that segfaults) can't crash the server 🔸 Vector search overhaul — QBit precision tuning, optimal bfloat16→int8 quantization, dimension strides for Matryoshka embeddings, random AdamR rotations, and a quantized codec with RaBitQ/TurboQuant built in 🔸 New regex engine — a hand-written library that compiles regular expressions to native code, more than 2x faster than RE2 and beating even vectorscan 🔸 Join acceleration — automatically use the right-hand side of a join as an index to filter the left, plus a new DPsub join-reordering algorithm; some queries run 8x faster or more 🔸 Dates before the French Revolution — DateTime range extended far beyond the old 20th–23rd-century limit, with all functions working across the full span 🔸 EXPLAIN ANALYZE + filtered projections — actually run a query and see per-stage timing, parallelism, and rows read; plus projections that store just the subset of rows matching a filter 🔸 More goodies — Iceberg manifest compaction, Delta Lake / Microsoft OneLake writes, UTM/MGRS geo coordinates, GeoJSON output, faster Arrow/ZSTD/LZ4/gzip, and the new groupFormat aggregate and eval/queryRunner table functions