# SchemaFill preprint reports up to 4.05× throughput for AI tool calls

_Published Wednesday, October 7, 2026 at 12:12 AM EDT · AI, Science · Latest · Tier 2 — Notable_

Researchers report that SchemaFill improved end-to-end throughput by up to 4.05× over token-by-token generation on the Glaive and BFCL benchmarks. The preprint presents a method for generating structured calls that let language models interact with external systems.

SchemaFill generates candidate argument values concurrently across fields and calls. The target model checks those candidates against the actual preceding output, commits only verified tokens, and supplies corrections when candidates disagree. This verification addresses dependencies that can make independently generated arguments differ from the target model's output. The reported improvement is a benchmark maximum.

## Sources

- [cs.LG updates on arXiv.org](https://arxiv.org/abs/2610.07086)

---
Canonical: https://techandbusiness.org/newswire/xRH4AJx0TfTXZvsMhItmY7
Published: 2026-10-07T04:12:37.362Z
Story chronology: 2026-10-07T04:00:00.000Z
Retrieved: 2026-10-07T06:07:12.874Z
Publisher: Tech & Business (techandbusiness.org)