14B open model on one RTX 4090 matches hosted frontier model on text-to-SQL; mnemiq open-sourced
ycombinator · x · 2026-09-09
After running 28 configurations, grading 25,000+ answers, and benchmarking against Snowflake and Databricks, the team found a 14B open model on a single RTX 4090 matched their hosted frontier model on text-to-SQL. Their takeaway: model size, vendor reputation, and published benchmark scores are poor predictors of performance on your own database. They've open-sourced mnemiq so you can test on your own data.
More from coding & agent
- Prompt optimizer GEPA lifts Meta Muse Spark 1.1 success from 22.2% to 100% while cutting queries to 0.3% — iamrobotbear · 2026-09-09
- Seroter's Daily Reading List: AI coding stacks, backlog burn-down and the future of software engineering — rseroter · 2026-09-09
- Google Cloud August AI Infra Roundup: gVisor Sandboxes on Ray, Filestore on Colossus — dl_weekly · 2026-09-09
- Grok Bot integrates with X, bundling free Starter API credits — xiaohu · 2026-09-09
- New Urbit library %opal turns every agent state face into a scry endpoint with one line — BLUECOW009 · 2026-09-09
- OpenClaw 2026.9.3 ships: 1,844 PRs, cloud repo work, live browser automation — steipete · 2026-09-09