OpenAI introduced GPT-6.2 Astra and Sol on November 17, 2026, enhancing AI capabilities for professional and everyday tasks. GPT-6.2 Astra, designed for complex work, scored 5.2 points higher than GPT-6 Astra on the Agents' Last Exam at less than a third of the cost while offering a 20% discount on input and output tokens.
These new models build on OpenAI's previous releases, including GPT-6 Astra and GPT-6.1 Sol, aiming to deliver frontier performance with greater efficiency and lower costs. The launch targets businesses and developers seeking scalable AI solutions, alongside individual users in ChatGPT Work.
The release emphasizes improved quality for demanding tasks, focusing on professional outputs, coding, and scientific research. GPT-6.2 Sol offers near-Astra intelligence for a fraction of the price, making advanced AI more accessible for diverse applications.
GPT-6.2 Astra Elevates Professional Workflows
GPT-6.2 Astra sets a new quality standard for AI in complex professional tasks, generating cleaner code and more professional documents, slides, and spreadsheets. It significantly outperforms previous models on benchmarks like Agents' Last Exam and Terminal-Bench-Science 0.1, offering improved performance at substantially lower costs due to better token efficiency.The model excels in creating professional outputs that adhere to industry standards in fields like law, finance, and accounting. For instance, on the Harvey LAB-AA v1.1 legal benchmark, GPT-6.2 Astra more than doubled Claude Fable 5.2’s pass rate at about one-sixth the cost.
Its capabilities extend to detailed computer use, interactive web design, and 3D modeling. For example, it can transform written briefs into interactive browser experiences, such as a playable pinball table or an explorable Colosseum model.
We evaluated GPT-6.2 Astra on turning dense information into readable briefs and infographics for both technical and non-technical audiences. Examples included trading briefs and diagrams of agentic workflows. GPT-6.2 Astra's documents were a significant step up from GPT-6 Astra. The writing and charts made it much easier to pick out the key points of a briefing, and the diagrams clarified complex systems by showing how the parts work together.
— Marquis Wong, Principal AI Engineer, IMC
How Does GPT-6.2 Sol Serve Everyday Users?
GPT-6.2 Sol is designed as an efficient and versatile model for everyday tasks, offering a balance of intelligence and cost-effectiveness. It is smarter and more efficient than GPT-6.1 Sol, allowing users to complete more tasks within the same paid plan, making it a go-to model for a broad range of applications.This model powers through multi-step agentic workflows and produces clear, natural writing. On Cognition’s FrontierCode 2.0, GPT-6.2 Sol achieved within 0.5 points of GPT-6.2 Astra at about one-fifth the cost. It also outscores GPT-6 Astra on the LifeVal evaluation for challenging everyday tasks.
The efficiency and intelligence of GPT-6.2 Sol make it suitable for tasks ranging from shortlisting kindergartens to planning meals for a family. This aligns with a trend toward more practical and accessible AI agents, as explored in discussions around whether AI Agents From Anthropic and OpenAI Aren't Killing SaaS, but Watch Out.
| Feature | GPT-6.2 Astra | GPT-6.2 Sol |
|---|---|---|
| Primary Use Case | Hardest professional projects, complex tasks | Efficient, versatile for everyday work |
| Agents' Last Exam | 5.2 pts higher than GPT-6 Astra | Not specified, but generally lower |
| Terminal-Bench-Science 0.1 | 9 percentage points higher than GPT-6 Astra | Not specified |
| Input Token Price (per million) | $8 | $2 ($2 per million input tokens, $10 for output) |
| Output Token Price (per million) | $40 | $10 |
What About Enhanced Security and Alignment?
OpenAI has integrated its strongest safeguards to date into GPT-6.2 Astra and Sol, implementing tightened restrictions and enhanced monitoring systems. Both models meet the Critical threshold for cybersecurity and the High threshold for biological and chemical capabilities under OpenAI's Preparedness Framework.Despite increased capabilities in finding vulnerabilities and writing secure code, OpenAI has launched these models with stricter safeguards. These measures reduce misuse risk, even if they restrict some legitimate vulnerability research and security testing workflows. Internal systems like 'Control Room' track research workloads and can halt model output when intervention is necessary.




