A 4B model fine-tuned with agentic RL beat Postgres's own planner by 1.81x, for about $1,200.

Rohan Bansal distilled 120 trajectories from GPT-6 Astra into a Qwen 3.8 4B base via LoRA (21.2M trainable of 4.66B), then ran agentic RL where the model emits PostgreSQL hints for join ordering and execution strategy, an agent harness executes the plan against four containerized Postgres instances, and reward comes from measured latency against the default plan. Training on the 13,646-query Cardinality Estimation Benchmark, validated on the 113-query Join Order Benchmark over IMDb with topologies verified distinct. Best-of-3 gave 1.81x geometric mean speedup and 44.7% summed latency reduction, 68 queries improving past 5% and exactly one regressing past it. $800 Lambda H100 rental, $400 API, code public at polyphilz/qorl. This is a usable template for reward-from-measured-execution against any optimizer you can instrument.