Chatbots & Virtual Assistants
Conversational AI for customer or general-purpose use. Ceux en tête vous en disent le plus sur leur fonctionnement — ce sur quoi ils ont été entraînés, où vont vos données et qui les a conçus.
Mis à jour le 11/19/2026
- 1ChatGPT· OpenAI
ChatGPT is OpenAI's AI chat assistant; ChatGPT Edu is the edition OpenAI offers universities for students, faculty, researchers and campus staff, run in a dedicated workspace with SAML SSO, SCIM, group permissions and an admin console. Students and staff use it to ask questions, draft and summarise text, analyse data and uploaded files, search the web, and build custom GPTs for a course or project. The vendor says ChatGPT Edu data and conversations are not used to train OpenAI's models by default, that retention windows can be set per workspace, and that eligible new Edu workspaces can store content at rest in Europe (EEA and Switzerland) and keep model inference on GPUs in Europe. OpenAI names GPT-6 as the flagship model available in ChatGPT Edu.
95% - 2Gemini· Google DeepMind
Gemini is a family of multimodal large language models developed by Google DeepMind, serving as the successor to LaMDA and PaLM 2. It is designed to process and generate text, code, images, audio, and video, enabling advanced reasoning, problem-solving, and task automation across diverse domains. Gemini models are available in different sizes (Ultra, Pro, Nano) to cater to various use cases, from enterprise applications to on-device processing. The models are integrated into Google's ecosystem, including Google Cloud, AI Studio, and consumer products like Bard (now rebranded as Gemini).
90% - 3Zendesk AI· Zendesk
Zendesk AI is an artificial intelligence-powered suite of tools integrated into the Zendesk customer service platform. It enhances customer support operations by automating responses, providing intelligent insights, and improving agent productivity. Key features include AI-driven chatbots (Answer Bot), intelligent triage, macro suggestions, sentiment analysis, and summarization of customer interactions. Zendesk AI leverages machine learning to analyze customer queries, predict intent, and offer personalized responses, thereby streamlining support workflows and improving customer experience.
90% - 4Claude 3 Opus· Anthropic
Claude 3 Opus is the most powerful model in the Claude 3 family, released by Anthropic in March 2024. It excels in complex reasoning, advanced coding, nuanced understanding, and multimodal tasks (e.g., analyzing images, documents, and charts). The model is optimized for high-stakes, high-impact applications and demonstrates state-of-the-art performance on benchmarks like MMLU, GPQA, and human preference evaluations. Claude 3 Opus was trained on a proprietary mix of publicly available internet data (up to August 2023), private datasets, and internally generated data. It underwent RLHF (Reinforcement Learning from Human Feedback) and Constitutional AI fine-tuning to align with principles like helpfulness, honesty, and harmlessness. The model is multimodal (text and vision) and multilingual. Safety evaluations show strong alignment, with low rates of harmful outputs and high resistance to jailbreaks. It was deployed under AI Safety Level 2 (ASL-2) protections.
90% - 5Devstral Small 2 (24B)· Mistral AI
Devstral Small 2 (24B) is a large language model (LLM) developed by Mistral AI, designed for advanced text generation, coding assistance, and problem-solving tasks. It is part of the Devstral family of models, optimized for efficiency and performance in enterprise and developer applications. The model is tailored for tasks such as code generation, debugging, natural language understanding, and technical documentation creation. It is positioned as a smaller, more accessible version of Mistral AI's larger models, making it suitable for deployment in environments with computational constraints.
90% - 6Gorgias AI Agent· Gorgias
Gorgias AI Agent is the generative AI layer of the Gorgias ecommerce helpdesk, used mainly by Shopify, BigCommerce, Magento and WooCommerce merchants. It answers shopper questions autonomously over email and chat (and is marketed for sales and proactive engagement as well as support), drawing on the merchant's Help Center, written Guidance, public URLs and live order and fulfilment data, and can carry out order actions through integrations depending on configuration. According to the vendor's AI Agent security FAQ, it runs on a mix of third-party LLMs (OpenAI and Anthropic) and in-house fine-tuned models, passes each draft through a second-model confidence check, and lets merchants define exclusion and handover topics; every AI answer is logged with the knowledge sources used. Typical users are DTC brand support teams that want to automate tier-1 tickets such as order status, returns and shipping questions. Rated medium. Under the EU AI Act this is a limited-risk system: a chatbot that interacts directly with consumers, so Art. 50 transparency applies, and Gorgias leaves AI disclosure largely to the deployer (chat messages are labelled 'Automated', email disclosure needs a merchant-configured signature). The data is ordinary consumer contact, order and payment-status data rather than special categories, and the assurance posture is reasonable (vendor-reported SOC 2 Type II since 2020 and HIPAA attestation, GCP hosting with EU residency for EU merchants, a DPA with 72-hour breach notice and SCCs). The open points that keep it above low are: the FAQ admits the AI Agent learns from interactions across brands; Anthropic is named as an LLM provider but is missing from the published sub-processor list; most sub-processors are US-based; and the agent can act on orders, not just reply. The rating would rise if merchants in health and wellness or similar verticals let it handle health-related questions (special category data), if it were used to make refund or fraud decisions with significant effect on shoppers, or if cross-brand learning were found to reuse identifiable data. It would fall with an ISO 27001 or ISO 42001 certificate, a complete sub-processor list and contractual opt-out from cross-customer learning.
90% - 7Microsoft Copilot· Microsoft Corporation
Microsoft Copilot (until 2026 named Microsoft 365 Copilot) is an AI assistant built into Microsoft 365 apps such as Word, Excel, PowerPoint, Outlook and Teams; Microsoft Copilot Chat is the chat version that work and school accounts get at no extra cost. Universities and schools use it to draft and summarise documents, emails and meetings and to answer questions; Microsoft says Copilot Chat is available to educators, staff and students aged 13 and over, is on by default for higher-education students, is off by default for 13-17-year-olds in K-12 tenants until an admin sets their age group, and is blocked for under-13s, while the full Copilot licence is sold to education organisations as an add-on. The vendor says Copilot combines large language models with content in Microsoft Graph that the user already has permission to see, and that prompts, responses and Graph data are not used to train foundation models. Microsoft says generated responses are not guaranteed to be factual and that users should review output before using or sharing it.
90% - 8C4AI Command A (03-2025)· Cohere
C4AI Command R+ (03-2025) is a large-scale, advanced language model developed by Cohere For AI. It is part of the Command R series, designed for enterprise-grade applications requiring high performance in reasoning, multilingual capabilities, and tool use. The model is optimized for retrieval-augmented generation (RAG) and workflow automation, making it suitable for complex business tasks such as customer support, content generation, and data analysis. It supports multiple languages and is built to handle long-context tasks efficiently.
85% - 9eduGenAI· SURF
eduGenAI is the generative-AI chat platform that SURF and Npuls are building for Dutch education (mbo, hbo, wo); it is in a pilot phase until the end of December 2026, and each institution decides which teachers, support staff and students get access. Users chat with a choice of language models, set up personas, and ground answers on uploaded documents; SURF says it can also be used for research and operational purposes, although institutions such as Utrecht University currently limit their pilot to education. The vendor says the platform offers open-source models that SURF hosts on its own hardware in a Dutch data centre (the SURF AI Hub) next to commercial models such as OpenAI GPT models via a SURF-managed Microsoft Azure environment, and that prompts and outputs are used only to generate answers, not to train models. SURF's DPIA found 12 low risks and recommends measures against over-reliance on the accuracy of generated text, and SURF tells institutions to check the compliance of each model they switch on themselves.
85% - 10Lyra 2.0· NVIDIA
Lyra 2.0 is a cutting-edge framework designed to generate persistent, explorable 3D worlds from a single image. It advances the state of the art in single-image 3D scene generation by enabling scalable, spatially persistent scene generation with real-time rendering capabilities.
85%
Comment nous avons fait notre sélection +
Nous attribuons à chaque IA une note sur 100 en fonction de ce qu'elle partage sur elle-même. Nous examinons cinq éléments :
- Qui d'autre est impliqué — vous indique-t-elle quelles autres entreprises traitent vos données ?
- Règles qu'elle respecte — a-t-elle été vérifiée au regard des lois et des normes ?
- Ce qu'elle indique aux utilisateurs — dispose-t-elle d'une politique de confidentialité, de conditions et d'engagements de sécurité clairs ?
- Comment elle fonctionne — explique-t-elle sur quoi elle a été entraînée et comment elle a été conçue ?
- Agir correctement — dispose-t-elle de règles sur un usage sûr et équitable ?
Une IA doit partager au moins la moitié de ces éléments pour figurer sur la liste. Nous classons par score. En cas d'égalité entre deux IA, la plus populaire l'emporte. Chaque entreprise ne peut apparaître qu'une seule fois par liste.
Nous obtenons ces informations à partir de ce que les entreprises partagent publiquement, de l'EU Open Source AI Index et de notre propre catalogue. Personne ne paie pour figurer sur cette liste.