Mistral introduces new APIs for content moderation and batch processing

France-based artificial intelligence company Mistral for content moderation released its new API. The API in question is the same API that powers moderation on Mistral’s Le Chat chatbot platform.

According to the information shared by the company; The API, which can be adapted to specific applications and security standards, is supported by the finely tuned model of Ministral 8B. Ministral 8B fine-tuned model, texts sexuality, hatred and discrimination, violence and threats, dangerous and criminal content, self-harm, health, finance, law ve personal It was trained to classify into nine categories: The model in question offers support in 11 languages, including English, French and German. Considering that competitors mainly focus on English, it is an advantage that Mistral also offers support in languages ​​such as Arabic, Chinese and Japanese. Mistral’s moderation API can be applied to raw texts or transcripts.

In his blog post, Mistral stated that the content moderation classifier uses the most relevant policy categories for effective guardrails. While Mistral claims that the moderation model is extremely accurate, he also acknowledges that it is a work in progress. So far we have encountered popular examples in this space, such as Jigsaw’s Perspective API and OpenAI’s moderation API. To get started using Mistral’s moderation API, the company shared: from documentation You can benefit from it.

Mistral has a batch processing option today. batch API Let us also point out that he introduced . Thus, the company can reduce the cost of models served through its API while processing high volumes of requests asynchronously. reduce by 25 percent aims. Anthropic’s affordable batch processing option that challenges OpenAI We told you about the Message Batches API. It’s worth noting that OpenAI and Google also offer batch processing options for their APIs. You can access Mistral’s batch API via La Plateforme.