OpenAI Unveils GPT-6.2 Astra and Sol, Two New AI Models
OpenAI Group PBC today announced the release of GPT-6.2 Astra, a frontier AI model designed for complex professional work, which it claims outperforms its predecessor at less than…
OpenAI Group PBC today announced the release of GPT-6.2 Astra, a frontier AI model designed for complex professional work, which it claims outperforms its predecessor at less than one-third of the cost. Also launched is an updated version of its everyday-use model, GPT-6.2 Sol. The two new models target enterprises and developers seeking frontier performance at scalable costs, and are said to deliver better results on demanding tasks while using fewer output tokens.
According to OpenAI, GPT-6.2 Astra set a new high on Agents' Last Exam, scoring 5.2 points higher than GPT-6 Astra, at less than one-third of the cost. It is also said to lead Anthropic PBC's Claude Fable 5.2 by 5.5 points, at just 5% of the cost. OpenAI positions GPT-6.2 Astra as the world's best model for scientists, with a score on Terminal-Bench-Science 0.1 that is 9 percentage points higher than GPT-6 Astra. In Harvey LAB-AA, the model's pass rate is more than double that of Fable 5.2, at roughly one-sixth of the cost.
Beyond benchmark score improvements, GPT-6.2 Astra is also trained to create slides, spreadsheets, and documents that more strictly adhere to the conventions used by professionals in fields such as law, sales, finance, and accounting. The model can match the layout, colors, and typography of slide templates in original files or screenshots, down to the exact font size. Early testers of the model include accounting AI startup Basis, which reported an average 50% reduction in time for its long-cycle accounting work, and up to a 73% reduction for its most complex tasks.
GPT-6.2 Sol is described by OpenAI as its go-to model for everyday work, and is said to allow users to accomplish more than with GPT-6.1 Sol under the same paid plan. According to OpenAI, the model scores within less than 0.5 points of GPT-6.2 Astra on Cognition AI Inc.'s FrontierCode 2.0 coding benchmark, at roughly one-fifth of the cost.
The release of GPT-6.2 Astra comes less than two months after OpenAI scrapped its planned October launch of GPT-6.1 Astra, following internal testing that found the model fell short in scope and authorization compliance. That decision was announced one day before the company's September DevDay event, where GPT-6.1 Sol was released on its own.
Like GPT-6 Astra, both GPT-6.2 models meet the "critical" threshold for cybersecurity under OpenAI's Preparedness Framework, and their enhanced capabilities have led the company to equip them at launch with stricter high-risk cybersecurity safeguards, which are said to also limit some legitimate vulnerability research. The company said it will work to bring the GPT-6.2 models to its Daybreak program for vetted cyber defenders. GPT-6.2 Astra is also the first externally released OpenAI model to reach the "high" threshold for AI self-improvement, a metric that measures a model's ability to accelerate AI research. OpenAI tracks this capability because models that help build future AI systems could outpace safeguards and human oversight; earlier this month, the company disclosed its Control Room system, deployed within its research infrastructure, which halts model outputs when intervention is needed.
The company noted that in its two alignment tests, the new models are more likely than GPT-6 Astra to omit mentioning obstacles that hinder their work, such as missing files. OpenAI said it has added instructions in Codex that significantly reduce such omissions and is continuing to investigate the behavior.
GPT-6.2 Astra and Sol are available starting today to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, with Enterprise and Edu access requiring workspace admin enablement. Developers can also access both models via the OpenAI API. GPT-6.2 Astra's API pricing is $8 per million input tokens, 40 cents for cached input, and $40 per million output tokens, reflecting a 20% discount on both input and output token prices compared to GPT-6 Astra. GPT-6.2 Sol maintains GPT-6.1 Sol's rates of $2 per million input tokens and $10 per million output tokens.
insigtX content is informational and educational, not investment advice.