About OpenGauntlet

Who publishes it

OpenGauntlet is published by D3velop. The work began as internal model-selection research for Imprynt and is shared publicly so builders can inspect the evidence instead of relying on vendor superlatives.

What is measured

The Words leaderboard directly measures local open-weight language models. Models receive the same multi-turn scenarios and instructions, then go through the same rubric and pairwise judging pipeline. Listening and Voices also publish smaller local experiments with their own disclosed recordings, listeners, hardware, methods, and limitations. Those labs are kept separate from the market surveys and are not population-wide rankings. The full conversational protocol is in the benchmark methodology.

What is surveyed

The Listening and Voices market catalogs, plus Turns, Runtimes, Memory, Hardware, Orchestrators, Quantization, the wider LLM field, and Utilities, are sourced comparisons—not claims that OpenGauntlet benchmarked every listed system. Entries are checked against primary papers, official documentation, source repositories, model cards, or measured receipts, and uncertainty remains visible.

Editorial standards

• Measured, vendor-reported, paper-reported, and inferred claims are labeled separately.
• Commercial-use and model-weight terms are checked separately from code licenses.
• Missing evidence is reported as missing; it is not converted into a score.
• Material corrections update the published data and provenance notes.
• No placement can be purchased and no vendor sponsors a ranking.

Corrections and contact

Every page links to its evidence. To challenge a claim or provide a stronger primary source, open a correction request. The public repository preserves the resulting change history.

Last reviewed: 2026-08-19. This page describes the standards applied to the currently published site.