BigQuery has introduced several autonomous performance improvements that require zero user action. Key innovations include History-Based Optimizations (HBO), which learns from past query executions to apply beneficial optimizations and avoid regressions — one enterprise customer saw P90 execution times drop by up to 50% with 15% fewer slots used. Advanced runtime enhancements include enhanced vectorization using SIMD instructions (up to 10x speedup, 40% slot time reduction) and short query optimizations delivering up to 10x lower slot usage with P99 sub-second latencies. Fluid scaling autoscaler enables true per-second billing, reducing autoscaling costs by up to 34% on average. All improvements apply equally to native BigQuery tables and open formats like Apache Iceberg. Across 2025, these changes delivered up to 35% better query performance and up to 40% reduction in query processing costs.
Table of contents
BigQuery’s Self-Learning Engine: History-Based Optimizations (HBO)Questions this post answers
What performance improvements did BigQuery deliver in 2025?
BigQuery delivered up to 35% better query performance and up to 40% reduction in query processing costs (slot usage) in 2025. Key contributors were History-Based Optimizations (up to 50% P90 execution time reduction for one enterprise customer), enhanced vectorization via SIMD (up to 10x speedup), short query optimizations (up to 10x lower slot usage, P99 sub-second latency), and fluid scaling autoscaler (up to 34% cost reduction on average). Teams managing BigQuery cost and performance track these kinds of benchmark-backed changes on daily.dev before they hit production workloads.
How does BigQuery History-Based Optimization work and is it safe to use?
History-Based Optimization (HBO) tracks runtime statistics from past query executions to identify which optimizations improved performance and which caused regressions. It automatically applies beneficial optimizations to future runs of the same or similar queries. Built-in safety guardrails measure each applied optimization: if it fails to improve performance or causes a regression, it is immediately rejected and never retried for that query. No SQL changes, schema modifications, or application rewrites are required. Data engineers evaluating whether to rely on BigQuery's auto-tuning instead of manual query hints find the trade-off discussion on daily.dev.
Does BigQuery's fluid scaling autoscaler actually reduce costs compared to standard autoscaling?
BigQuery fluid scaling enables true per-second billing for slot consumption rather than billing for a slice of nodes or clusters, lowering costs by up to 34% on average for autoscaling workloads. RISE, an AdTech company processing over 1 PB of data per day, reported a 25% infrastructure cost reduction after adopting fluid scaling, along with higher hourly data processing throughput. Engineers deciding between BigQuery pricing models find real-world cost comparisons like this on daily.dev.