September 4 News.OpenAI release GPT-6 Astra, for end-to-end work on complex reasoning, programming, computer operations, research and document creation。
The model is open to clients in Trusted Access Program, and API and ChatGPT Plus, Pro, Business and Enterprise users will be granted access over the next few days。
GPT-6 Astra supports 1.05 million token context windows and 12.8 million token maximum output, and adds the capability of a walk tool to call, to receive additional instructions in operation, to adjust the strength of reasoning in dialogue。
OpenAI indicates that the model can perform multistep tasks in browsers and specialized software。
In terms of safety, OpenAI rated Astra as "Critical" for cybersecurity, and "High" for biological and chemical capabilities, and limited some of its advanced network security capabilities to Daybreak Blue。
With respect to pricing, the GPT-6 Astra standard model provides $10 per million input token charge, $1 in cache input, $12.5 in cache and $50 in output。
Enter a request in excess of 272,000 token at a rate of 2 times the Cache and 1.5 times the output rate; 50% for the Batch and Flex mode prices, and two times the corresponding price for Fast Mode。