Finance edition

OpenAI releases GPT-6.1 Sol. For now, the company is holding back GPT-6.1 Astra for safety reasons. OpenAI says Sol performs almost as well for coding, using a computer and professional work. Sol costs about one fifth as much as Astra. That is around $2 per million input tokens and $10 per million output tokens. Tokens are small pieces of text that a model processes. You can use Sol through the API, ChatGPT Work and Codex. The API lets your own software use the model. For finance teams, this could significantly change the estimated cost of AI tasks. But a lower price per unit does not tell you the total cost of a workflow. The bill also depends on how much text and work the system processes. So do not compare model prices alone. Compare the cost of doing the same task, too. Work out the costs using your own usage before putting a saving in the budget. With an API, an organisation pays for each token. If the price per million tokens falls, the same input and output cost less. But the total bill depends on two things. How many tasks do you run? And how much text does each task process? So a cheaper model only brings costs down if you compare like with like. Have both models do the same tasks. OpenAI says Sol performs almost as well on the tasks mentioned. The source gives no independent comparison. It also gives no results for financial tasks. Astra is being held back for safety reasons. So do not assume Sol offers the same quality or carries the same risks for every process. A lower price per task may also mean extra checking work. That can happen if you need to review the result more often.

Source →