A new runtime named 'l' has emerged, designed to execute k4, q, and q SQL code natively. Built on a modern engine, 'l' prioritizes memory bandwidth by treating compressed vectors as the primary unit of execution. Its primitives operate directly on encoded bytes, avoiding the need to decompress and rebuild full arrays, a strategy aimed at leveraging hardware capabilities that are often starved by data movement.

'l' maintains full compatibility with existing k4, q, and q SQL syntax, idioms, and code. Users can run their current tables, dictionaries, and time-series data without modifications. The runtime features end-to-end column compression, enabling direct execution on compressed vectors to minimize data transfer. It also offers transparent SIMD, threading, and hardware offloading, automatically selecting the most efficient execution path (scalar, SIMD, threaded, or offloaded) without requiring explicit annotations from the developer.

Performance benchmarks indicate significant speedups compared to reference implementations. For instance, 'l' achieved a 3.57x geomean speedup on the TSBS benchmark and a 3.16x speedup on the db-benchmark per-query geomean. This performance gain is attributed to its approach of moving fewer bytes and enabling primitives to run across wider SIMD plans and multiple threads. A plain-text mailing list is available for users discussing development with 'l'.