# LLM Reliability Benchmark dataset changelog

All dates are measurement or first-publication dates, not invented revisions.

## v1.0 — 2026-11-12

Initial public release, published with the LLM Reliability Benchmark 2026 page.

- 12 open chat models, reached on Cloudflare Workers AI’s free tier.
- 40 tests in five categories (prompt injection, structured output, instruction following, refusal calibration, grounded answers).
- Deterministic grader only (no LLM judge). Suite, grader and runner included.
- Licence: CC BY 4.0. Credit “Litmiq LLM Reliability Benchmark 2026” and link to the benchmark page.
- Canonical files also remain at /assets/benchmark/ (unversioned aliases of v1.0).
- No later version has been published as of 13 November 2026.
