ChatGBot
ModelosFuncionesCreador de appsPreciosFAQ
Iniciar sesiónEmpieza gratis
← Volver al blog2 de octubre de 2026

Tavus Griffin: The AI That Nearly Passed a Video Turing Test

By ChatGBot · October 2, 2026

On October 1, 2026, Tavus unveiled Griffin, which it calls the first Human Interaction Model. It is built to hold real-time, face-to-face video conversations, and in one study nearly half of the people talking to it believed it was a real human. Here is what Griffin is and why the numbers caused a stir.

ClaudeGPTDeepSeekQwenGLM

Every top AI in one app

GPT, Claude, DeepSeek, Qwen and GLM — one subscription.

Try ChatGBot →

What is Griffin?

Griffin is a full-duplex, video-to-video model that talks with you face to face in real time. Instead of generating a clip after you finish speaking, it adapts its words, timing, tone, expressions, and movements to what is happening in the moment. It can react to interruptions, know when to laugh or lean in, and even wave at someone who walks into the room. If you care about talking rather than typing, see our guide to an AI voice assistant.

The numbers behind the hype

  • 48% thought it was human. In a live study, 48% of participants believed they were speaking with a real person, up from just 2.4% with Tavus's previous Phoenix-4.5 setup.
  • Near-human on benchmarks. On NVIDIA's VideoFDB benchmark it scored 3.83 out of 5, close to the 3.92 human reference and well above the 2.80 of the next published system.

Can you use it yet?

Not quite. A limited Griffin-Lite is restricted to select trusted testers, and Tavus says broader customer access waits on safety work, with no firm general-release date. So the headline results are impressive, but this is early access, not a public launch.

Why it matters

Griffin is a sign of where conversational AI is heading: not just text and voice, but lifelike video you can talk to naturally. It arrives alongside a busy season of releases like GPT-6.1 Sol and Gemini 4 Argon. For now, the most practical way to use today's best models is to keep them in one place and pick the right one per task.

Frequently asked questions

What is Tavus Griffin? A real-time, video-to-video Human Interaction Model that holds face-to-face conversations and reacts naturally to what is happening.

Did Griffin pass the Turing test? In one live study, 48% of people believed they were talking to a human, a large jump over Tavus's earlier models, though it is not a formal universal Turing test.

Can I try Griffin? Only a limited Griffin-Lite is open to select trusted testers; general availability has no firm date yet.

Is Griffin a chatbot like ChatGPT? No. It is a video avatar model for face-to-face conversation, not a text assistant. For text and voice chat across top models, ChatGBot gives you GPT, Claude, DeepSeek, Qwen and GLM in one app.

Una suscripción. Varios modelos de IA líderes.

Empezar gratis

Blog

  • DeepSeek V4.1-Flash: Features, Context Window and October Update
  • GPT-6 Pro Parallel Agents: What the New ChatGPT Update Does
  • Best AI in 2026: The Top Models and How to Pick
  • Gemini Embedding 2: The Multimodal Search Model Getting Buzz
ChatGBot

Una suscripción para todos los modelos de IA líderes: chat, imágenes, búsqueda web, documentos, Canvas y el Creador de apps.

support@chatgbot.ai

Producto

ModelosFuncionesCreador de appsPreciosFAQIniciar sesión

Legal

Términos de usoPolítica de privacidadPolítica de Reembolso
BlogDeepSeek V4.1-Flash: funciones, ventana de contexto y actualización de octubreAgentes paralelos de GPT-6 Pro: qué hace la nueva actualización de ChatGPTBest AI in 2026: The Top Models and How to PickGemini Embedding 2: el modelo de búsqueda multimodal que está dando que hablarTodos los artículos

© 2026 ChatGBot. Operado por 42 Dijital Yazılım Limited Şirketi.

Chatgbot es una interfaz de IA independiente. No está afiliada, respaldada ni patrocinada por OpenAI, Anthropic, Google, xAI, DeepSeek, Qwen, GLM u otros proveedores de modelos. Los nombres de los modelos y las marcas registradas pertenecen a sus respectivos propietarios. La disponibilidad de los modelos puede variar según el plan, la región y el acceso del proveedor.