What AI Models is Hugo Using?

Hugo uses several best-in-class AI models to power its features. Depending on your needs, you can choose the model that fits your use case best.

Available Models

OpenAI GPT 5 & GPT 5.1

The latest generation of OpenAI models. GPT 5 and GPT 5.1 offer excellent general-purpose performance across all Hugo AI features, including conversation summarization, reply suggestions, and knowledge base search.

Anthropic Claude 4.5 Sonnet & Haiku

Claude 4.5 Sonnet provides high-quality, nuanced responses and is great for complex tasks. Claude 4.5 Haiku is a lighter, faster variant ideal for quick interactions where speed matters most.

Mistral Large & Medium

Mistral models are built in Europe and offer strong multilingual capabilities. Mistral Large is suited for demanding tasks, while Mistral Medium provides a good balance of performance and speed. Mistral is a great choice if you need EU data residency, as the models are hosted in Europe.

Crisp Internal Model (Self-Hosted by Crisp)

This model is developed and hosted entirely by Crisp on Scaleway infrastructure in Paris, France. It is the best option if your organization requires maximum privacy and compliance with EU regulations (such as GDPR), since no data leaves Crisp's own servers.

Google Gemini Technical Preview

Google Gemini is available as a technical preview. It provides strong performance and is continuously being improved by Google. Note that this model is still in preview and may evolve over time.

Pricing

Most models are equivalent in terms of pricing. You can switch between models without worrying about significant cost differences.

Privacy & EU Compliance

If privacy or EU compliance is a priority for your organization, we recommend using one of the following models:

• Crisp Internal Model — Fully hosted by Crisp on Scaleway in Paris. No data is sent to any third-party AI provider.
• Mistral — European-built models with data processed in Europe.

Both options ensure that your data stays within the EU and complies with European data protection regulations.

Built-in Models Trained by Crisp

In addition to the third-party AI models listed above, Hugo also relies on built-in models that were trained internally by Crisp. These models run alongside the main AI model you choose and handle critical tasks automatically:

• Guardrails / Hallucination Mitigation: A dedicated model checks Hugo's responses to detect and prevent hallucinations, ensuring that answers remain accurate and grounded in your knowledge base content.
• Spam Fighting: A built-in model identifies and filters spam messages in real time, so that Hugo and your team can focus on genuine customer conversations.
• Escalation Decisions: A specialized model determines when a conversation should be escalated to a human agent, ensuring that complex or sensitive issues are handed off at the right moment.

Updated on: 01/02/2026