Google's DualSQL Uses Multi-Agent RL for Text-to-SQL, 8B Model Beats Prior 32B Systems
omarsar0 · x · 2026-09-22
DualSQL from Google and colleagues trains Text-to-SQL agents with multi-agent RL: one agent links questions to tables/columns, another writes SQL, both sharing the same model weights and trained in a single run.
- Agents can query the database via three tools while reasoning.
- To prevent multi-agent RL collapse, the authors add rollout guardrails and a new reward, robust execution match, for more accurate SQL correctness judging.
- Trained on only 3,755 examples, DualSQL-4B hits 68.0% execution accuracy on BIRD dev, matching earlier 7B models; DualSQL-8B reaches 71.1%, beating previous single-model systems with 32B parameters.
More from coding & agent
- Microsoft's Agensh Scales Multi-Agent Systems to 1,024 Agents Without a Central Orchestrator, Boosting Test-Pass Rate to 55% — andrew_n_carr · 2026-09-23
- Viral demo claims 'GPT-6' can drive browser Paint to draw, unverified — alexcovo_eth · 2026-09-23
- Parallel coding agents merge cleanly and silently break every test — RunAI_Coder · 2026-09-23
- Getting phone-captured text into your computer: OCR, vision models, agentic pipelines — silenceimpaired · 2026-09-23
- Free Bots: a persistent 3D city where AI agents work, earn, buy land and build houses — Daniel_Farinax · 2026-09-23
- Game dev looks like the programming field most resistant to AI — how much is it actually used? — marktenenholtz · 2026-09-23