JA EN
shiichan

GPT-6 Astra Has Arrived! Work in ChatGPT Work and Codex Just Got a Major Upgrade

OpenAI News ★★★★★ View Markdown
#openai#ai#ai-safety#security

Hi everyone, this is Shii-chan! OpenAI just announced a new model called GPT-6 Astra, and it's already generating a ton of buzz.

OpenAI News openai.com

What was announced?

This news comes from OpenAI's News page, introducing the new model GPT-6 Astra. It's already available in ChatGPT Work, Codex, and the API, and OpenAI says it's state-of-the-art at computer use, browsing, professional work, software engineering, cybersecurity, and science. The pitch is that teams can now take on the toughest professional work with more speed, accuracy, and judgment.

The story so far

Until now, putting AI to work at a business usually meant preparing your data, redesigning workflows, and building custom integrations before you'd see any value. Astra is meant to change that. Inside ChatGPT Work and Codex, it can write code and operate the same everyday applications people already use, even ones without an API. That means teams can plug it into their existing workflows from day one, without a big engineering lift up front.

What changes

Just days into rollout, OpenAI says customers are already putting Astra to work on things like:

  • Optimizing GPUs
  • Spotting discrepancies in financial statements
  • Producing more on-brand slide decks

Internally, OpenAI rolled Astra out weeks before launch. Its developer and marketing teams used Astra and Codex to turn three hours of multicamera footage into a "GPT-6 Astra Developer First Impressions" video, which racked up over 550,000 views in just four days. The engineering team also used Astra to track down a memory-allocation bottleneck that was slowing down Codex sessions in a test environment. Switching allocators delivered 25x lower turn latency, at the cost of roughly 30% higher peak memory use. On top of that, Astra is better at matching a company's voice, templates, and design standards, so first drafts land closer to something a team can actually use.

Dive Deep

Cost efficiency and pricing

Astra is trained to finish tasks in fewer tokens and with fewer retries, which means less rework and a lower cost per task. OpenAI says this puts it at the majority of the cost-efficiency frontier on professional-work and coding evaluations, including Terminal Bench 4.0 and the Artificial Analysis Intelligence Index. Pricing starts at $10 per million input tokens and $50 per million output tokens.

Safety and control

OpenAI describes Astra as its most aligned model yet, with stronger adherence to human intent and authorization. On an internal computer-use safety benchmark, testing scenarios like exposing confidential information, oversharing a dashboard, or deleting data, Astra produced unintended outcomes 89% less often than GPT-5.6 Sol, and 74.7% less often than Claude Fable 5.1.

Organizations also get new admin controls, including the ability to:

  • Restrict access to approved websites and desktop applications
  • Manage uploads and downloads
  • Control browsing history

ChatGPT Work and Codex also add confirmation policies that require approval before consequential actions, plus automated review of potentially unsafe or unauthorized tool calls, so teams can start with a limited setup and expand access over time.

For ChatGPT Desktop, OpenAI is also launching new enterprise plugins built on its latest browser-use capabilities, including:

  • Oracle Analytics
  • Power BI (a Microsoft Fabric service)
  • Navan
  • Avalara

Astra is also the first model to reach the "Critical" cybersecurity capability threshold under OpenAI's Preparedness Framework. In response, OpenAI says it strengthened protections against misuse and unauthorized actions, training Astra to respect safety and security boundaries, improving its resistance to jailbreak attempts, and deploying automated checks to block harmful responses. Zero Data Retention is also available for eligible API customers on supported endpoints, subject to approval.

Wrap-up

  • The new GPT-6 Astra model is now available in ChatGPT Work, Codex, and the API
  • It can operate everyday apps directly and write code even without an API
  • One internal case saw turn latency drop up to 25x
  • Pricing starts at $10 per million input tokens and $50 per million output tokens
  • Safety benchmarks show far fewer unintended outcomes, alongside new enterprise controls and plugins

If you want AI woven deeply into daily work, or you're a team weighing security and access controls as you roll out AI, this launch is worth a closer look!

Related posts