The request names ToxiGen and HateBERT, which are research models for hate-speech detection rather than products; none of the candidate sources describes or hosts them directly. The closest fits are sources offering text moderation or classification: OpenAI's moderation action 111 and Replicate's ability to run open-source ML models via API 1020. Hugging Face-style BERT model access is shown in NVIDIA's TAO skill that integrates HuggingFace models 919.

A question back: Do you want to run the ToxiGen and HateBERT models yourself (for example from Hugging Face), or call a hosted moderation service? Which platform or environment will you run them on?

1. 1

   OpenAI APIopenai.comverified and signed

   Its moderation action checks text against OpenAI's rules on hate, violence and similar harm, a comparable hate-speech classification option 111.

   WebsiteDocs

   official:API · OpenAI APISDK · openailibrary · OpenAIskill · babysit-prSDK · openai+163 community tools built for it

   for agents: classify_text · analyze_sentiment · createModeration

2. 2

   Replicatereplicate.comverified and signed

   Runs open-source machine learning models through a cloud API, so community models such as ToxiGen or HateBERT classifiers could be run there 1020.

   WebsiteDocsPricingGet started

   +12 community tools built for it

   for agents: find-models · run-models · publish-models

3. 3

   NVIDIA TAO Toolkitnvidia.comverified and signed

   Its skill integrates and fine-tunes HuggingFace computer vision, VLM and LLM models on NVIDIA GPUs, and it is a route to fine-tuning BERT-family text models 919.

   WebsiteDocsPricingGet started

   +81 community tools built for it

   for agents: tao-port-huggingface-model · tao-finetune-huggingface-model · tao-train-grounding-dino