OpenAI unveils cost-effective GPT-6.1 Sol model

OpenAI has unveiled the new GPT-6.1 Sol language model, designed specifically for writing code, automating computer operations and carrying out complex sequences of actions. The release replaces the previous GPT-6 Sol version.

The developers say the new model comes close to the flagship GPT-6 Astra in performance while costing users significantly less. The algorithm has already been integrated into the ChatGPT Work workspace and the Codex platform, while third-party software developers can access it through the API. However, this version is not used for standard everyday conversations in the regular ChatGPT. OpenAI reports improvements in programming, computer control and professional tasks. The model is expected to perform better at writing and debugging code, analyzing documents and handling scenarios that require commands to be executed step by step.

The key advantage of the new version over Astra is its cost-effectiveness. API pricing for GPT-6.1 Sol is set at $2 per million input tokens, $0.10 for cached tokens and $10 for output tokens. For GPT-6 Astra, meanwhile, these rates are $10, $1 and $50 for the same amount of information. Therefore, the cost of standard data input and output with Sol is reduced fivefold, while processing repeat requests from the cache is ten times cheaper.

At the same time, according to the developers, the performance tradeoff is minimal. On the DeepSWE platform, GPT-6.1 Sol delivered results close to those of Astra and outperformed the previous GPT-6 Sol version by 6.4% even with less in-depth task analysis. In AutomationBench tests, the new model surpassed Anthropic Opus 5.5 by 2.2% at medium reasoning settings, while the computations cost approximately three times less.

In addition, OpenAI is working to integrate Ultrafast mode. The company says this option will increase data generation speed in Codex by up to eight times compared with standard mode, although an exact release date has not yet been disclosed.