09:43Machine-translated from Azerbaijani by AIRead in Azerbaijani
OpenAI unveils more powerful GPT-6.1 Sol AI model
OpenAI showcased the GPT-6.1 Sol model at DevDay 2026. According to the company, the new AI model delivers results close to those of the more powerful GPT-6 Astra in programming, computer use and professional tasks. The model is an updated version of GPT-6 Sol, which was released just 1 week ago, and uses fewer tokens.
In addition to programming and fixing errors, GPT-6.1 Sol outperforms the previous model at understanding documents. It also handles tasks consisting of several stages more effectively. In simple reasoning mode, the share of responses containing factual errors fell from 11.4% to 7.7%. In other modes, its error rate differs from that of GPT-6 Astra by no more than 1.9%. The model also states more clearly when it cannot do something or lacks the necessary capabilities. OpenAI originally planned to showcase the GPT-6.1 Astra model at the DevDay event. However, it halted the model's release because of safety concerns. In internal tests, the model displayed more deceptive behavior and continued carrying out tasks without the user's permission. The company says GPT-6.1 Sol did not try to evade automated safety checks. The model complies more closely with imposed restrictions, reports when the search tool is not working and is less likely to produce unauthorized results.
With GPT-6.1 Sol, 1 million input tokens cost $2 (approximately 3 manat). The price of 1 million cached input tokens is $0.10 (approximately 0.17 manat). Output costs $10 (approximately 17 manat). A token is a small piece of text that the model reads or generates.
Standard input and output prices are 1/5 of the price of GPT-6 Astra. The price of cached input is approximately half that of the previous GPT-6 Sol. The model is now available in ChatGPT Work and Codex to Plus, Pro, Business, Enterprise and Edu subscribers. For now, it cannot be used in the regular chat section. The always-on personal AI assistant Dots was also showcased at the event. This agent works in connected applications, completes tasks based on available information and learns from user feedback. Dots runs in the background on a secure cloud computer. The agent is gradually becoming available to users of the Pro and Business Premium plans.
OpenAI also announced the Pro 500 plan, priced at $500 per month (approximately 850 manat). This package is considerably more expensive than the company's current $200-per-month plan (approximately 340 manat). The package's Ultrafast mode can generate 300 tokens per second in the Codex and Work applications. That is 8 times faster than standard mode. The API version will also operate 6 times faster than the regular speed. The company has also created a separate space called ChatGPT Space for teamwork. Employees can work on the same project there and share conversations and common information needed for their work. Dots assistants can also access that information and carry out tasks in line with the team's work.
In addition to programming and fixing errors, GPT-6.1 Sol outperforms the previous model at understanding documents. It also handles tasks consisting of several stages more effectively. In simple reasoning mode, the share of responses containing factual errors fell from 11.4% to 7.7%. In other modes, its error rate differs from that of GPT-6 Astra by no more than 1.9%. The model also states more clearly when it cannot do something or lacks the necessary capabilities. OpenAI originally planned to showcase the GPT-6.1 Astra model at the DevDay event. However, it halted the model's release because of safety concerns. In internal tests, the model displayed more deceptive behavior and continued carrying out tasks without the user's permission. The company says GPT-6.1 Sol did not try to evade automated safety checks. The model complies more closely with imposed restrictions, reports when the search tool is not working and is less likely to produce unauthorized results. With GPT-6.1 Sol, 1 million input tokens cost $2 (approximately 3 manat). The price of 1 million cached input tokens is $0.10 (approximately 0.17 manat). Output costs $10 (approximately 17 manat). A token is a small piece of text that the model reads or generates. Standard input and output prices are 1/5 of the price of GPT-6 Astra. The price of cached input is approximately half that of the previous GPT-6 Sol. The model is now available in ChatGPT Work and Codex to Plus, Pro, Business, Enterprise and Edu subscribers. For now, it cannot be used in the regular chat section. The always-on personal AI assistant Dots was also showcased at the event. This agent works in connected applications, completes tasks based on available information and learns from user feedback. Dots runs in the background on a secure cloud computer. The agent is gradually becoming available to users of the Pro and Business Premium plans. OpenAI also announced the Pro 500 plan, priced at $500 per month (approximately 850 manat). This package is considerably more expensive than the company's current $200-per-month plan (approximately 340 manat). The package's Ultrafast mode can generate 300 tokens per second in the Codex and Work applications. That is 8 times faster than standard mode. The API version will also operate 6 times faster than the regular speed. The company has also created a separate space called ChatGPT Space for teamwork. Employees can work on the same project there and share conversations and common information needed for their work. Dots assistants can also access that information and carry out tasks in line with the team's work.