Quail: open-source AI-SQL engine hits 1B+ input tokens/min on a single H100
sh_reya · x · 2026-09-27
Researcher shreya, in collaboration with Modal, released Quail, an open-source AI-SQL engine. By planning SQL queries and LLM inference together, it reaches 1B+ input tokens/min on a single H100 for a single query.
The core idea: AI-powered data operators create a new class of inference workload, and Quail "pushes LLM inference closer to the data/query layer" — unlike the Jevons-style shift of decision-making toward the model layer, it co-designs query execution with inference scheduling to massively speed up AI queries over large datasets.
More from Infra
- Bought a second RTX 3090 for local AI but see no benefit yet—what are dual-3090 setups for? — BlueSky4200 · 2026-09-27
- AI's boom is the power of a trillion dollars in spending, not exponential growth — kareem_carr · 2026-09-27
- Raspberry Pi says memory costs jumped 3.7x in six months as AI datacenters hog supply — julsimon · 2026-09-27
- Google's Suncatcher to test TPU radiation tolerance on prototype satellite via SpaceX — julsimon · 2026-09-27
- YC's AI Office Hours: Wafer hits 379ms vs Cerebras 674ms, 44% lower latency — ycombinator · 2026-09-27
- Cerebras built a 20x faster Grok bot: dinner booking done in 22 seconds — Sethwinterroth · 2026-09-27