OpenAIModelsSafety and security
OpenAI releases GPT-6.2 Astra and Sol
Astra beats Claude Fable 5.2 on two tough tests at a fraction of the cost, while Fable leads on 3D design and the Artificial Analysis index. Pricing and access.

Cover of the Instagram post.
Sources: OpenAI, ChatGPT
OpenAI’s strongest model for cyber
OpenAI calls GPT-6.2 Astra its most capable model ever for cyber. On the internal ExploitBench test it scores 22 points above GPT-6 Astra: 53.5% versus 31.5%.
Astra and Sol both meet the Critical cybersecurity threshold under OpenAI’s Preparedness Framework, so they launch with tighter safeguards for higher-risk cyber help. The chart was measured with those safeguards turned off.
Ahead of Fable 5.2 on Agents’ Last Exam
On this tough agent test, Astra scores 78.3% while Claude Fable 5.2 scores 72.8%. OpenAI says Astra costs just 5% as much per task: $2.89 versus $60.05.
OpenAI’s footnote says the Fable 5.2 result excludes nine tasks that used a fallback model.
Also ahead on Terminal-Bench-Science
Astra scores 77.1% on this scientific terminal test, 9 points above GPT-6 Astra at less than a third of the cost per task.
In OpenAI’s run, Claude Fable 5.2 scores 73.3%, but with a different agent harness. Anthropic reports 74.6% for Fable 5.2 in Claude Code.
Fable 5.2 still leads in 3D design
BenchCAD 2.1 Core tests whether models can build 3D parts from engineering drawings and fit them into assemblies. Claude Fable 5.2 leads at 83.5%, with Astra just 1 point behind at 82.5%.
Astra costs about a tenth as much per task: $5.43 versus $50.77. The more efficient Sol scores 73.1%.
Fewer tokens, better results
OpenAI says both models deliver better results on demanding tasks with fewer output tokens. On GDPval v2, Astra scores 6.2 points above GPT-6 Astra while using 22% fewer output tokens. The garden planning site in the video is one of OpenAI’s web design examples.
Car dynamics on a mountain road
OpenAI asked Astra to visualize the chassis dynamics of a car on a mountain road. Astra built the 3D model, wrote a Python solver for the interaction between the tires and suspension, and animated the results in Blender.
Pricing and access
Astra is the frontier model for the hardest projects, while Sol is the more efficient model for everyday work. OpenAI says most of its employees use Sol as their default.
In the API, Astra costs $8 per million input tokens and $40 per million output tokens, 20% less than GPT-6 Astra. Sol costs $2 and $10.
Both models are rolling out in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and are available in the API from day one.
Not first on every test
On the Artificial Analysis Intelligence Index chart on OpenAI’s own page, Claude Fable 5.2 leads with 64.6. Astra is second at 62.1, followed by Gemini 4 Carbon (60.0), Claude Opus 5.5 (59.1) and Sol (57.3).
So Astra beats Fable 5.2 on two tests but trails on the overall index. Its real edge is doing the same work at a much lower cost.
Sources
5 sources
About this story
- Written by
- Yiğit Kasap · JARRUS
- Story date
- On the site
- since Nov. 18, 2026
- @jarrus.tech posted this story on Nov. 18, 2026.
- Short link
thejarrus.com/en/gpt-6-2-astra-sol- Report an error
- Spotted an error in this story? [email protected] · Instagram