OpenAI has announced the introduction of two AI models, GPT-6.2 Astra and GPT-6.2 Sol, designed to address a broad range of professional and routine tasks.

The technology company says the new models are now available to Plus, Enterprise, Pro, and Edu subscribers in ChatGPT Work and Codex, with developers able to access them via the OpenAI API.

Access deeper industry intelligence

Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.

Find out more

The models are also available with Ultrafast for Pro 500 and, for the first time, Pro 200 subscribers, beginning with GPT-6.2 Sol.

According to OpenAI, GPT-6.2 Astra is intended for complex and demanding projects in professional and scientific domains, while GPT-6.2 Sol is aimed at the efficient management of everyday workflows.

GPT-6.2 Astra is positioned as a higher-performing option at a lower cost than previous versions, reportedly achieving superior results in industry benchmarks.

On Agents’ Last Exam, which assesses multi-step professional tasks, GPT-6.2 Astra exceeded the performance of GPT-6 Astra by 5.2 points, and outperformed Claude Fable 5.2 by 5.5 points, reportedly at a fraction of the associated costs.

The company states that GPT-6.2 Astra is suited for technical applications including scientific research, legal work, engineering design, and finance.

In the Terminal-Bench-Science 0.1 test, focused on scientific tasks in a terminal environment, GPT-6.2 Astra reportedly scored nine percentage points higher than GPT-6 Astra.

The model also led other benchmarks measuring biological data interpretation and experiment simulation.

Pricing details provided by OpenAI place GPT-6.2 Astra’s API rates at $8 per million input tokens, $40 for output tokens, and $0.40 for cached input, reflecting a 20% discount compared to GPT-6 Astra and a 60% reduction on cached inputs.

GPT-6.2 Sol retains GPT-6.1 Sol’s rates at $2 for input, $0.10 for cached input, and $10 for output per million tokens.

Both models, according to OpenAI, aim to deliver improvements in cost-efficiency and token usage.

For legal, finance, and consulting fields, OpenAI highlights improvements in accuracy and adherence to professional templates.

In Artificial Analysis’s AA-AnalystAgent tests, GPT-6.2 Astra correctly answered 62.5% of questions across five attempts when processing complex financial reports.

The models also bring enhancements in engineering, design, and cybersecurity.

According to the company, both GPT-6.2 Astra and GPT-6.2 Sol meet the Critical threshold for cybersecurity under its Preparedness Framework, with additional safeguards implemented to limit risks associated with higher-risk security tasks.

In the announcement, OpenAI said: “Developers increasingly rely on coding agents to write production software, even as more capable agents lower the barrier to finding and exploiting vulnerabilities.”

The company reports that new training signals have been added to improve defensive capabilities for developers.

OpenAI further describes strengthened controls around the training and deployment of GPT-6.2 Astra and GPT-6.2 Sol.

Safeguards have been designed to monitor model outputs and restrict dual-use cases involving biological or cybersecurity concerns.

Both models reportedly meet the High threshold for biological and chemical capabilities, as well as for AI self-improvement, as defined by the company’s Preparedness Framework.

Recently, California Attorney General Rob Bonta filed a lawsuit against OpenAI, claiming the company did not implement adequate safety protocols and oversight during the development and testing of its advanced AI models.