22:05Machine-translated from Azerbaijani by AIRead in Azerbaijani
OpenAI Unveils Cost-Effective GPT-6.1 Sol Model
OpenAI has unveiled the new GPT-6.1 Sol model, designed for programming, computer use, and multistep tasks. According to the company, the new model approaches the more powerful GPT-6 Astra in capabilities but costs significantly less to use via the API.
GPT-6.1 Sol replaces the previously introduced GPT-6 Sol model. The new model is currently used within ChatGPT Work and Codex and is also available to developers through the API. However, GPT-6.1 Sol is not intended for everyday conversations in the standard version of ChatGPT.
The new model has been optimized for programming, performing various computer operations, and professional work. It delivers better results in writing and debugging code, analyzing documents, and carrying out complex tasks that require several consecutive steps. OpenAI specifically presents GPT-6.1 Sol as a model that balances Astra-like performance with a lower price for complex coding, computer use, and professional work.
One of the main differences is the price. For GPT-6.1 Sol, 1 million input tokens cost $2 via the API, cached input tokens cost $0.10, and output tokens cost $10. For GPT-6 Astra, the same figures are $10, $1, and $50, respectively. This makes Sol's standard input and output tokens 5 times cheaper than Astra's, while its cached input tokens are 10 times cheaper.
According to data published by OpenAI, despite the price difference, the performance gap between the two models is relatively small across a range of complex tasks. For example, GPT-6.1 Sol achieved a result close to Astra's on the DeepSWE test and improved its score by 6.4% compared with the previous GPT-6 Sol. The company also says the new model outperformed Anthropic's Opus 5.5 by 2.2% on the Automation Bench test in medium reasoning mode, while completing the task at approximately three times lower cost.
A faster Ultrafast mode is also being developed for GPT-6.1 Sol. According to OpenAI, this mode can increase token generation speed in Codex by up to 8 times compared with standard mode. An exact release date for the mode has not yet been announced.
GPT-6.1 Sol currently supports a context window of 1.05 million tokens and a maximum of 128,000 output tokens via the API. The model also offers reasoning options at low, medium, high, xhigh, and max levels.