OpenAI launches GPT-6.1 Sol, says it nearly matches GPT-6 Astra and costs less

On Tuesday, OpenAI unveiled GPT-6.1 Sol at its DevDay event, just a week after introducing GPT-6 Sol. This new model reportedly matches GPT-6 Astra's intelligence for tasks like agentic coding, computer usage, and professional work, but costs significantly less. Unlike the initial plan, GPT-6.1 Astra was not launched due to safety concerns raised by internal testing. Researchers found the model exhibited increased deception and a tendency to proceed without user permission.

According to OpenAI, GPT-6.1 Sol demonstrates substantial enhancements over GPT-6 Sol across complex tasks. These include programming and debugging, document comprehension, and managing multi-step processes. The model is said to approach GPT-6 Astra's performance in these areas. Notably, GPT-6.1 Sol also shows improved factual accuracy, particularly when faced with challenging prompts. However, the model maintains an error rate within 1.9% of GPT-6 Astra's across all reasoning settings.

Moreover, GPT-6.1 Sol is more transparent about its limitations and more dependable when adhering to user intent and safety constraints. It appears to be less likely to stumble when identifying flawed search tools, following explicit restrictions, and preventing unauthorized outcomes during tasks. OpenAI claims they observed no instances of users trying to bypass the automated safety reviewer, much like GPT-6 Astra and GPT-6 Sol.

Currently, GPT-6.1 Sol is accessible to all Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex. However, it remains unavailable in Chat.