OpenAI has launched GPT-6 Astra, a high-end AI model designed for complex end-to-end tasks, advanced reasoning, coding, computer use and professional workflows, with rollout planned across ChatGPT and API platforms.
OpenAI has taken another major step forward in model development with the release of GPT-6 Astra, its newest high-end model.
The model was officially launched on September 03, 2026 and is being introduced as the company's most capable model for difficult, end-to-end work.
Unlike earlier generations that mainly focus on giving better answers, Astra is designed around a broader idea: helping users complete complicated work from beginning to end.
That means understanding a goal, working through several stages, using available tools, handling changes in instructions, and producing a finished result.
The release comes shortly after the GPT-5.6 family, making Astra a particularly significant development in OpenAI's rapidly changing model lineup.
Company leaders hail the Astra model as a "generational leap" that moves the industry into the AGI (Artificial General Intelligence) era.
Loading post from https://x.com/OpenAI/status/2095595741528125780…
Astra can interact directly with computer desktop environments, operating systems and web browsers.
It executes multi-step professional tasks, such as managing software pipelines, filling out complex forms, and coordinating applications independently.
The model works best when paired with the ChatGPT Desktop App to leverage its native system-level automation features.
Astra sets a new standard for AI reasoning and logic across several highly advanced benchmarks.
It achieves 98% on FrontierMath Tier 4, helping mathematicians tackle open research problems.
This model scores 99.9% on the ARC-AGI-3 metric, demonstrating near-perfect abstract reasoning capabilities.
Records a 100% score on ExploitBench, highlighting its precision in analysing digital security frameworks.
Astra can write tests and deploy entire codebases or complex scripts from a single prompt.
It creates interactive 3D assets and environments directly in professional software like Blender, such as historical reconstructions.
Developers use Astra to write and manage multi-agent simulations inside Unreal Engine.
Astra is the first model rated as the "Critical" risk threshold under OpenAI's Preparedness Framework.
It can autonomously discover and exploit zero-day vulnerabilities across hardened networks and operating systems without human intervention, triggering high-level security containment protocols during deployment.
It was released on March 05, 2026, and was designed specifically around professional work.
It combined advances in reasoning, coding, and tool-based workflows and brought capabilities to documents, spreadsheets, and presentations.
It also introduced an important idea for ChatGPT: the model could provide a plan while working on a complex response, allowing users to redirect the work before it was finished.
Astra builds on this general direction, but it is positioned as a higher capability level, particularly for long-running, computer-based, end-to-end work.
This model arrived shortly before Astra and was already positioned as OpenAI's flagship model for demanding work.
Sol introduced deeper reasoning settings and an ultra mode designed to coordinate multiple agents for particularly difficult tasks.
It also showed strong performance in coding, professional analysis, computer use and science.
Astra therefore should not be viewed as a completely different model from GPT-5.6. It is better understood as the next generation in that direction.
GPT-5.6 Sol established a strong balance between capability, efficiency, and cost across the Terra and Luna models.
Astra moves the focus toward the upper end of capability, where the priority is handling the most difficult work possible.
The rollout began on September 03, 2026, for trusted organisations and participants in OpenAI's cyber application program.
Over the following days, it will launch for ChatGPT Plus, Pro, Business, and Enterprise tiers, alongside developer access via the OpenAI API and Amazon Web Services (AWS).
It is a premium, resource-intensive architecture priced at $10 per 1 million input tokens and $50 per 1 million output tokens, making it roughly 2.5 times more expensive than its predecessor, GPT-5.6 Sol.
GPT-6 arrives at a time when expectations around ChatGPT are changing quickly.
Astra stands out because it combines those capabilities in one high-end model.
It has a very large context window, multiple reasoning levels, advanced coding abilities, stronger computer-user performance, research capabilities, and the ability to create professional documents and other materials.
At the same time, its increased capabilities have forced OpenAI to pay much closer attention to security monitoring and safeguards.
