GPT-6 Just Rendered a Universe from Text

Matthew Berman recently unveiled After Hours, an infinitely explorable metropolis generated by an AI named Astra. This GPT-6 model, described by OpenAI as a "generational leap," showcased its capacity to construct an entire dynamic world from astonishingly minimal input. The project's genesis from just two text prompts signals a profound shift in generative AI capabilities.

Every element within After Hours—from towering skyscrapers to bustling pedestrians and even falling rain—renders exclusively as ASCII characters. This creates a unique, living reality, where intricate 3D environments emerge from a tapestry of text. Users navigate a continuously populating landscape, witnessing a vibrant, animated city out of seemingly simple characters.

The generative nature ensures infinite exploration; as users traverse the city, Astra dynamically expands the world, populating new streets and scenarios. Berman's showcase reveals a sophisticated AI capable of not only interpreting abstract commands but also manifesting complex, interactive virtual spaces with unprecedented efficiency, challenging our understanding of digital world creation.

Berman's infinitely explorable "After Hours" city, built by Astra using only two prompts, directly showcases the capabilities of OpenAI's recently released GPT-6 Astra. OpenAI unveiled Astra with a limited preview on September 3, 2026, and general availability on September 4, 2026. The company positions this model as a "generational leap," touting it as their "most intelligent and aligned model" yet.

OpenAI asserts Astra achieves state-of-the-art performance across a wide spectrum of complex tasks, demonstrating unparalleled proficiency. Its claimed expertise spans critical domains including:

• Computer use
• Browsing
• Software engineering
• Scientific research
• Professional work

Specifically relevant to Berman's demonstration, Astra shows exceptional aptitude for generating intricate video game scenes and constructing dynamic virtual environments.

Beyond Astra's confirmed prowess, industry chatter also swirls around an unconfirmed, even more potent model: GPT-6 Sol. Rumors suggest Sol completed a "zero-shot 3D-city task" in merely 15 minutes, processing 60,000 tokens to create a complex environment without explicit prior training for that specific output type. This speculative model, if real, underscores the breakneck pace of development in generative AI, hinting at capabilities that could soon dwarf Astra's impressive feats and redefine digital world creation.

The 2026 landscape has seen a boom in text-to-environment AI, with notable contenders like Google DeepMind's Genie 3 and Autodesk's Wonder 3D. These models have advanced the generation of intricate 3D assets and contained scenes from text prompts. Yet, Astra's "After Hours" showcase fundamentally redefines this frontier, moving beyond isolated creations.

A critical distinction sets Astra apart: its ability to generate a complete, interactive, and continuously expanding 3D world, not merely a single object. While Genie 3 and Wonder 3D excel at rendering individual assets or finite scenes—like a detailed car or a specific room—Astra builds entire generative environments. It dynamically populates infinite pathways, allowing for seamless, unbounded exploration, as demonstrated by Matthew Berman's two-prompt city.

This breakthrough immediately revolutionizes industries demanding rapid, immersive world-building.

• Gaming studios can prototype expansive open-world environments in minutes, drastically cutting development cycles.
• Film pre-visualization gains unprecedented agility, enabling directors to explore full virtual sets from minimal descriptive text.
• Architectural simulation benefits from instant, explorable urban or interior designs, transforming conceptualization workflows into interactive experiences.

No longer do creators spend months meticulously crafting individual assets or designing intricate levels. Astra's "After Hours" demo, which built a dynamic, infinitely explorable city from a mere two prompts, signals a profound paradigm shift in virtual world-building. This acceleration moves development from a labor-intensive endeavor to a prompt-driven act of instantaneous creation.

This constrained, prompt-to-world methodology extends beyond just large labs and their flagship models. The very concept of a complex, interactive cyberpunk ASCII city, like "After Hours," illustrates how minimal input can yield rich, explorable environments. This focused generation proves a wider trend in creative AI, emphasizing ingenuity and high-level conceptualization over brute-force asset production, democratizing complex world design.

AI is evolving beyond a simple tool for generating static assets into a true collaborative partner. Models like GPT-6 Astra now interpret high-level intent to construct entire complex, interactive systems. The generative nature of "After Hours," continuously populating an infinite world with walking characters and dynamic elements like rain, showcases AI's capacity to co-create living, evolving virtual spaces. Developers now orchestrate emergent realities, engaging AI in a profound new creative dialogue.

What is the 'After Hours' project shown in the video?

It is an interactive, infinitely explorable 3D city rendered entirely with ASCII characters. According to creator Matthew Berman, it was generated by OpenAI's GPT-6 Astra model from just two text prompts.

What AI model created the ASCII city?

The model is named Astra, which is identified as part of the GPT-6 family of models recently released by OpenAI in September 2026.

What makes the 'After Hours' city special?

Its core features are its unique ASCII aesthetic, its infinitely generative nature that builds the world as you explore, and its dynamic elements like rain and walking characters, creating a sense of a living environment.

Is text-to-3D world generation a new technology?

While text-to-3D object generation has been improving, the ability to generate entire navigable, dynamic environments from simple text prompts is a recent breakthrough, with companies like Google (Genie 3) and OpenArt also pushing the frontier in 2026.