# OpenGauntlet

> Independent results and research for locally hosted conversational AI, speech recognition, and text-to-speech systems.

## Primary pages

- [Conversational AI leaderboard](https://opengauntlet.com/): Measured local open-weight language models.
- [Open-weight LLM survey](https://opengauntlet.com/llms.html): Sourced comparison of the wider open-weight model field.
- [Methodology](https://opengauntlet.com/methodology.html): Scenarios, judging, ratings, calibration, and reproducibility.
- [Speech-to-text measurements](https://opengauntlet.com/asr.html): Scoped local ASR experiment with disclosed method and receipts.
- [Speech-to-text survey](https://opengauntlet.com/asr-survey.html): Sourced comparison of 106 ASR systems.
- [Turn-taking research](https://opengauntlet.com/turns.html): Sourced VAD and endpointing system comparison.
- [Inference engine research](https://opengauntlet.com/runtimes.html): Sourced LLM serving/runtime comparison, including a first-party vLLM vs SGLang benchmark.
- [Hardware reference](https://opengauntlet.com/hardware.html): GPU and hardware VRAM/bandwidth reference, model-agnostic.
- [Orchestrator research](https://opengauntlet.com/orchestrators.html): Sourced voice-agent framework and platform comparison.
- [Memory research](https://opengauntlet.com/memory.html): Sourced vector database and agent-memory system comparison.
- [Quantization research](https://opengauntlet.com/quantization.html): Sourced quantization toolchain comparison.
- [Utilities research](https://opengauntlet.com/utilities.html): Sourced speaker-verification, wake-word, and speech-enhancement comparison.
- [Text-to-speech measurements](https://opengauntlet.com/tts.html): Scoped local TTS blind-listening experiment with disclosed method and receipts.
- [Text-to-speech survey](https://opengauntlet.com/tts-survey.html): Sourced comparison of 168 TTS systems.
- [TTS comparison guide](https://opengauntlet.com/tts-guide.html): How to interpret TTS evidence.
- [TTS legal guide](https://opengauntlet.com/tts-legal.html): Licensing, consent, and legal-risk research.

## Machine-readable data

- [Leaderboard JSON](https://opengauntlet.com/data/leaderboard.json)
- [Methodology JSON](https://opengauntlet.com/data/methodology.json)
- [ASR research JSON](https://opengauntlet.com/data/asr.json)
- [ASR lab JSON](https://opengauntlet.com/data/asr-lab.json)
- [Turns research JSON](https://opengauntlet.com/data/turns.json)
- [Runtimes research JSON](https://opengauntlet.com/data/runtimes.json)
- [Hardware reference JSON](https://opengauntlet.com/data/hardware-catalog.json)
- [Orchestrators research JSON](https://opengauntlet.com/data/orchestrators.json)
- [Memory research JSON](https://opengauntlet.com/data/memory.json)
- [Quantization research JSON](https://opengauntlet.com/data/quantization.json)
- [Open-weight LLM survey JSON](https://opengauntlet.com/data/llms.json)
- [Speaker verification research JSON](https://opengauntlet.com/data/speakerid.json)
- [Wake word research JSON](https://opengauntlet.com/data/wakeword.json)
- [Speech enhancement research JSON](https://opengauntlet.com/data/enhancement.json)
- [TTS research JSON](https://opengauntlet.com/data/tts.json)
- [TTS lab JSON](https://opengauntlet.com/data/tts-lab.json)