OpenAI's GPT-6 Astra Is Shockingly Good at Almost Everything
OpenAI launched GPT-6 Astra on September 3, a significantly more expensive model that excels at spatial reasoning, mechanical tasks, and code generation but shows weaker writing ability than its predecessor. Early testers demonstrated the model's capabilities by rapidly creating complex 3D environments, playable games, and other visual projects during the launch weekend.
Why It Matters
Astra represents a notable shift in AI capabilities toward visual and interactive tasks rather than language proficiency, suggesting the field's priorities are evolving toward embodied intelligence and practical applications. The model's pricing increase combined with performance tradeoffs highlights ongoing questions about which capabilities are most valuable to developers.
Key Facts
- Release date and pricing: September 3, $10 per million input tokens and $50 per million output tokens (2.5x the cost of predecessor)
- Computer use benchmark: 72.6% of desktop tasks completed on OSWorld 2.0 at roughly 40 minutes per task
- 3D city recreation: Developer SuSu rebuilt Hangzhou in a web browser using Three.js in 24 minutes
- Game development: Rishi Prasad built a browser-based multiplayer shooter with 12-person lobbies in one day
- Cybersecurity capability: First OpenAI model rated at threshold for finding unknown software flaws
OpenAI's latest model, released at the start of September, marks a departure from the company's previous focus on language performance. Priced at five times the rate of earlier versions, Astra distinguishes itself through capabilities in visual understanding, spatial reasoning, and automated task execution rather than conventional text-based metrics.
Early adopters quickly demonstrated the model's prowess in 3D environments and game development. One developer generated a detailed recreation of Manhattan within Unreal Engine by processing the layout street by street, while another produced a navigable 3D model of a Chinese city with landmarks and interactive features in under half an hour. Game designers leveraged Astra to create everything from isometric-view games with character-driven narratives to browser-based multiplayer shooters complete with server architecture and voice communication, with some projects completed in a single day.
The model's ability to interact directly with desktop environments—manipulating mouse and keyboard inputs to complete real-world tasks—represents a significant advancement in agentic AI capabilities. Testing data showed Astra completing ordinary computer tasks at rates exceeding previous versions, and it achieved security relevance by identifying previously unknown software vulnerabilities independently.
However, the model comes with notable tradeoffs. Multiple testers noted that Astra's writing quality dropped measurably compared to its immediate predecessor, with performance benchmarks showing substantial declines on professional language tasks. This divergence suggests OpenAI has optimized the system for different priorities, trading general language ability for spatial reasoning and mechanical task execution.
The concentrated burst of capability demonstrations during launch weekend underscored where Astra excels and where it stumbles. Developers found it transformative for visual and interactive projects but less suitable for text-centric applications, establishing clear boundaries around its practical applications.
Keep Reading
Tracking cocoa may be just the beginning for PwC, Merck, Hashgraph provenance system
Xiaomi’s wide foldable promises more power than Samsung’s
First Xiaomi, then the world: why Arm might give phone gaming a huge graphics boost
