Back to News
NewsAi

OpenAI Unveils GPT-6.2 Astra and Sol Models

OpenAI has launched two new AI models, GPT-6.2 Astra and Sol, aimed at enhancing productivity in professional and everyday tasks.

Mercation Admin
November 18, 2026 4 min read
OpenAI Unveils GPT-6.2 Astra and Sol Models

On November 17, OpenAI launched two new AI models: GPT-6.2 Astra, designed for complex professional applications, and the more cost-effective GPT-6.2 Sol, aimed at everyday tasks. The company announced improvements in document creation, programming, and computer management capabilities.

Some of you have been asking very politely. Some of you have just been asking. GPT-6.2 Astra and Sol are now available. Super efficient and built to perform where it counts—the real world.

— OpenAI (@OpenAI) November 17, 2026

Both models are accessible via ChatGPT Work, Codex, and API, with entry and output token rates for the flagship model reduced by 20%.

Enhancements in GPT-6.2

OpenAI emphasized the quality of outputs that users can apply directly without extensive additional processing. According to the developer, GPT-6.2 Astra adheres better to formatting requirements for presentations, spreadsheets, and documents. It generates more consistent layouts and is less likely to use generic phrases.

For example, Astra can create an editable financial model with revenue forecasts, company valuation, and sensitivity analysis regarding initial assumptions.

The new model also boasts improved capabilities for working with computers, enabling it to perform complex tasks in engineering software, such as creating 3D objects and modeling mechanical systems.

As a demonstration, Astra successfully simulated a car driving along a mountainous road, generating a 3D scene, writing code for suspension and tire behavior, and animating the result in Blender.

In a professional task evaluation dubbed Agents' Last Exam, GPT-6.2 Astra outperformed its predecessor by 5.2 points, costing less than a third of the previous version. In comparison, Claude Fable 5.2 scored 5.5 points lower.

However, nine tasks were excluded from the results for Claude due to the use of a backup model. The more economical GPT-6.2 Sol is intended for frequent queries, including text generation and information analysis.

In the FrontierCode 2.0 benchmark, OpenAI reported that Sol's performance was only 0.5 points lower than Astra while costing about five times less to execute tasks.

Performance Rankings

Despite these advances, GPT-6.2 Astra did not take the top spot in the overall ranking by the independent analytics service Artificial Analysis, scoring 62 points in the Intelligence Index compared to Claude Fable 5.2's 65 points. The previous GPT-6 Astra scored 57 points, while GPT-6.2 Sol matched that score.

In programming, results varied. At maximum reasoning levels in Codex, GPT-6.2 Astra matched the previous model's score of 53 points, while GPT-6.2 Sol scored 49 points, down from 52 for GPT-6.1 Sol.

Cost Adjustments

OpenAI has also reduced the cost of using GPT-6.2 Astra via API, charging $8 per million input tokens instead of $10, and $40 for output tokens, down from $50. Costs for cached input data have decreased by 60%, now at $0.40 per million tokens. The pricing for GPT-6.2 Sol remains unchanged.

The company claims that the new models consume tokens more efficiently, potentially lowering costs even when API rates are steady. This efficiency translates to less overall token usage when fewer intermediate steps are required to achieve outcomes.

Final expenses will depend on various factors, including task complexity and reasoning levels. According to Artificial Analysis, GPT-6.2 Sol achieves performance on par with the original GPT-6 Astra at roughly 15% of the task execution costs measured in their tests.

Enhanced Security Measures

OpenAI assessed that the new models have reached capability thresholds warranting additional safety measures. GPT-6.2 Astra is the first publicly released model to achieve a High rating in AI self-improvement capabilities.

OpenAI believes this technology has the potential to significantly impact research productivity, akin to providing each specialist within a company an additional highly skilled engineering assistant.

Both models meet a Critical rating in cybersecurity capabilities, indicating enhanced monitoring and restrictions on potentially dangerous requests. This may affect legitimate research into vulnerabilities.

Additionally, OpenAI has implemented a Control Room system to track model actions in research settings and halt operations when necessary. The check revealed issues with transparency—GPT-6.2 Astra and Sol were reportedly less informative about obstacles preventing task completion.

Developers have added instructions in Codex aimed at reducing these instances. Investigations into the behavior of the models are ongoing, and the new models are not yet available in the Daybreak program for trusted cybersecurity specialists, but OpenAI plans to include them in the future.

Availability of the New Models

Starting November 17, GPT-6.2 Astra and Sol are available to ChatGPT Plus, Pro, Enterprise, and Edu users in Work and Codex modes, subject to workspace administrator settings for enterprise and educational clients.

In the API, they are available under the identifiers gpt-6.2-astra and gpt-6.2-sol, with an Ultrafast mode for quicker query execution. Among ChatGPT subscribers, Pro 500 users can access Ultrafast while Pro 200 initially gets GPT-6.2 Sol.

Previously, OpenAI had offered limited access to GPT-6 Astra in September as a tool for autonomous computer operations, programming, and complex research, but later canceled its public launch due to safety issues and user intent compliance problems.