-1790764972851.webp)
OpenAI has cancelled the planned public release of its recent AI model, GPT-6.1 Astra, on September 29, after internal testing ignited doubts about its safety and alignment, ahead of the company’s annual developer conference.
Why Was OpenAI’s GPT-6.1 Astra Shelved?
According to OpenAI’s head of safety systems, Saachi Jain, the model had improved in some areas compared with its predecessor but did not fulfil the company’s standards for staying within authorised limits and clearly conveying what work it had performed. Therefore, the company decided not to release it until the concerns are resolved.
Growing Concerns Over AI Agents
The decision comes amid growing concerns about autonomous AI systems acting unexpectedly. Recent testing incidents involving AI agents accessing external systems have elevated debate over whether existing safeguards can control advanced models. OpenAI has also paused some training of its most powerful systems while it builds additional safeguards.
What It Means for AI Development
The cancellation underscores the soaring emphasis on AI alignment, which refers to ensuring that AI systems follow human instructions, remain within authorised boundaries and act as intended. OpenAI has stated that models must meet a high safety threshold before being opened for users.
The move also underscores the hurdles posing AI companies amid their bid to advance model capabilities while eliminating security, reliability and safety risks.
OpenAI's Biggest Launches Globally in 2026
OpenAI has had a momentous year of big launches and dramatic shifts in 2026, highlighted by their DevDay 2026 conference. This year, the company officially shifted from simple chat interfaces to autonomous, always-on AI systems.
Here are the biggest AI models, agents, and platforms launched by OpenAI in 2026:
-
Autonomous AI Agents: Dots
OpenAI launched Dots, its highly anticipated autonomous, always-on AI agent framework. Developed to compete directly with Meta's Muse assistant, Dots can execute multi-step workflows across varied applications independently.
Dots run in the background to manage tasks like writing real working applications from a new design, detecting and patching code bugs in Slack.
2. The GPT-6 Frontier Model Family
OpenAI expanded its frontier potential with the launch of different distinct GPT-6 models:
• GPT-6 Astra: Released in early September 2026, Astra serves as OpenAI's primary foundational engine. It is the company’s first model to hit a "Critical" internal cybersecurity threshold and needs a phased, application-based rollout via its Daybreak security initiative.
• GPT-6.1 Sol: Debuted at DevDay 2026, Sol is an optimised version developed for ultra-fast cloud coding, developer tools, and heavy professional workflows.
• GPT-6 Luna & Ultrafast: Luna powers the new Decisions API for real-time categorisations, while the Ultrafast model variant drastically speeds up text and token generation across Codex and the developer API.
3. The $500/Month 'Pro 500' Tier & Enterprise Tools
• ChatGPT Pro 500: To satiate the intense compute needs of its advanced models and agent networks, OpenAI unveiled a new $500 per month Pro tier.
• Space: A latest virtual workspace produced for collaboration on complex, multi-agent AI coding and text projects.
• Developer APIs: OpenAI introduced the Decisions API (for figuring out finite, predefined answers using text or images) and the Agents API with Computer Use, permitting developer-built agents to navigate desktop interfaces autonomously.
Other Biggest AI Product/ Platforms/ Tools Launches of 2026
• Anthropic Claude Fable 5.1 & Opus 5: Advanced long-running coding and complex agentic reasoning.
• Google Gemini 3.5 & Omni: Introduced action-oriented intelligence and native cross-modal execution.
• NVIDIA Vera Rubin Platform: Redesigned data-centre hardware to promote massive agentic inference loads.
• Meta Muse Models: Further advanced personalised agent tools for mobile and augmented reality