OpenAI Launches GPT-6.1 Sol With Near-Astra Power At Lower Cost

OpenAI has introduced GPT-6.1 Sol at DevDay, pitching it as a model that gets close to the capabilities of its flagship GPT-6 Astra while costing far less to run. That distinction matters more than a benchmark headline. A useful model can reach many more products when developers can afford to put it behind everyday tasks, rather than reserve it for their hardest prompts.

In its announcement, OpenAI says Sol is built for agentic coding, computer use and professional work. In practical terms, those are jobs that require a model to work through several steps, use tools, inspect the result and correct itself. A coding assistant might read a repository, propose a change and run checks. An office assistant might work through a spreadsheet or a browser-based workflow. Neither becomes dependable merely because a model scores well in a test, but these are the uses OpenAI wants Sol to support.

The price is the clearest part of the story. OpenAI lists standard API input at $2 per million tokens and output at $10 per million tokens, with cached input at $0.10 per million tokens. It says that puts Sol at about one-fifth of Astra’s standard price. The cached-input rate is especially relevant for applications that repeatedly send the same instructions or background material, although a real bill will depend on how much context and output a service uses.

OpenAI describes Sol as nearly matching Astra on several measures of coding, computer use and professional work. Those are the company’s own evaluations, not a guarantee that Sol will perform equally well on every customer’s tasks. Developers still need to test their particular prompts, tools, latency and failure cases before swapping models in a production system. A cheaper answer is not automatically a better one if it needs more attempts or more human correction.

The release adds a middle option to OpenAI’s model lineup. Astra remains the premium choice for the most demanding work, while Sol is intended to make more capable AI practical at scale. That could affect the economics of customer support, software development and internal business agents. When a task involves thousands of calls a day, even a modest difference in price per call can decide whether a feature ships at all.

Sol is available through the API and across ChatGPT Plus, Pro, Business, Enterprise and Edu plans, according to OpenAI. Availability does not mean every user will see identical limits or every connected application will switch models immediately. Organisations with compliance requirements should also check the product’s controls and terms rather than assume that a general ChatGPT rollout is the same as an approved enterprise deployment.

There is a broader competitive point here. OpenAI is not alone in trying to make stronger models cheaper. Chinese AI labs and other Western providers have been pressing prices down, while businesses increasingly ask for measurable returns rather than another impressive demo. As our look at GPT-6 Astra noted, the frontier model attracts attention; a less expensive model that businesses can use every day may have the wider effect.

The test for GPT-6.1 Sol will come after the launch excitement. If developers find it reliable enough to handle complex workflows without turning every step into an expensive supervision exercise, the lower price could put more capable agents in ordinary software. If not, the savings on tokens may be swallowed by retries and human review. OpenAI has made its case. Customers now have to measure it against their own work.