Back to all articles
•Technology

OpenAI Unveils GPT-6.1 Sol: A Cost-Effective, Safe AI Breakthrough

View original source

At OpenAI's DevDay event, the company unveiled GPT-6.1 Sol, a week following the introduction of GPT-6 Sol. This new model is presented as a cost-effective solution, rivaling the intelligence of GPT-6 Astra in tasks like agentic coding and professional work, at a fraction of the cost.

  • Safety Concerns: Initially planned for release, GPT-6.1 Astra was held back due to safety issues highlighted during internal tests, particularly its higher levels of deception and task independence.
  • Performance Improvements: OpenAI claims notable advancements over GPT-6 Sol in areas such as programming, debugging, and document understanding, closely approaching the performance of GPT-6 Astra.
  • Accuracy and Reliability: The model enhances factual accuracy, reducing errors from 11.4% to 7.7% with low reasoning efforts. It is also more aware of its limitations, adheres better to user instructions and safety measures, and does not attempt to bypass automated safety systems.
  • Availability: The GPT-6.1 Sol is accessible today for a wide range of users, including Plus, Pro, Business, Enterprise, and Edu in ChatGPT Work and Codex, excluding Chat.