Export
The link always opens the latest version of this page.
Ask an AI about it
Open in ChatGPTA new chat that reads this entry from the site Open in ClaudeThe same, in ClaudeShare the link instead of a file, and they will always see the latest version.
Files are a snapshot of today. The live page has anything newer.
OpenAI Astra
Internal OpenAI model demonstrating advanced AI reasoning capabilities
- Category
- frontier language model 1 source
OpenAI Astra is a frontier language model 1 developed by OpenAI 2 in pre-launch form 3. The model, also referred to as GPT-6 Astra 4, demonstrates advanced reasoning capabilities with reported performance on security benchmarks.
Performance Characteristics
Astra achieved 100% accuracy on ExploitBench 5 and demonstrated arbitrary code-execution capabilities, zero-day exploitation, and privilege-escalation exploit development. The model reportedly conducts attacks at a higher rate than GPT-5.6 Sol and GPT-5.5, while maintaining a 0% scope-exceeding rate compared to 48% for GPT-5.6 Sol 5. On ExploitGym, the model scored 42.4% 5. It ranked 57 on the Artificial Analysis Intelligence Index.
Deployment
Astra is accessible through multiple platforms. These include ChatGPT Plus, Pro, Business, and Enterprise tiers 5, the OpenAI API 5, and cloud services including Amazon Web Services Bedrock 5 and Microsoft Azure. The model is also available through OpenRouter. Access operates on a subscription-based model.
The model supports integration with desktop applications, CRM systems, and ad platforms, and functions as a cloud-based API.
Note: All claims derive from single-source reporting and should be treated as preliminary information pending independent verification or official OpenAI documentation.
Facts 2
| Attribute | Value | Confidence | Sources | Cited |
|---|---|---|---|---|
| Vendor | OpenAI
also reported: Enovis (low)
|
low | 3 | 2 |
| Category | frontier language model | low | 1 | 1 |
Affected version 7
- pre-launch version
- conducts attacks at higher rate than GPT-5.6 Sol and GPT-5.5
- GPT-6 Astra
- 0% scope-exceeding rate vs 48% for GPT-5.6 Sol
- 100% on ExploitBench; arbitrary code-execution capabilities; zero-day exploitation; privilege-escalation exploit development
- 100% on ExploitBench; 42.4% on ExploitGym
- 57 on Artificial Analysis Intelligence Index
Version 1
Platform 10
- desktop applications, CRM systems, ad platforms
- cloud-based API
- OpenRouter
- Responses API, Codex, Azure, Bedrock
- subscription-based access
- ChatGPT Plus, Pro, Business, Enterprise; OpenAI API; Amazon Bedrock
- ChatGPT Plus, Pro, Business, Enterprise; API
- ChatGPT Plus, Pro, Business, Enterprise; OpenAI API; AWS
- ChatGPT Plus, ChatGPT Pro, ChatGPT Business, ChatGPT Enterprise, OpenAI API, Microsoft Azure, Amazon Web Services Bedrock
- Large Language Model
Neighbourhood
Affected by 1
Detects 6
Developed by 1
Involved in 4
Mitigates 1
Part of 10
Parts 5
Superseded by 1
Supersedes 2
Official links
Frequently mentioned together 15
Sources 206
- Managing frontier AI risk: From training to deployment · OpenAI
- An internal OpenAI Astra model solved 10 major open math and CS problems · Hacker News (YC)
- The AI Security Problem Is Bigger Than the Hugging Face Breach · HackerNoon
- Suspicious Minds: Unmonitorable AI Undermines Accountability · Lawfare
- OpenAI launches GPT-6 Astra, its first model to cross a critical cybersecurity threshold · CSO Online
5 of 206 sources shown · Pro unlocks the full list