Just a day after OpenAI Group PBC said it’s unwilling to release its latest model GPT-6.1 Astra on concerns over its safety, it has released a newer, almost as powerful model called GPT-6.1 Sol, which promises to deliver Astra-like performance at a much lower cost.
The new model, announced at OpenAI’s DevDay event today, is confusing because of the way it seems to make a mockery of OpenAI’s model naming convention, which became much more confusing during the summer when it first debuted GPT-5.6. Back then, it announced three versions of that model – Sol, Terra and Luna. Sol was the most powerful version, Terra sat in the middle, and Luna was the lower-cost, lightweight option for when costs and speed matter more than accuracy.
Though that originally made sense, things became more confusing when the company moved on to GPT-6 and gave it the “Astra” moniker. It said Astra is the most powerful version of the model, and shortly afterwards released two others in Luna and Sol. Terra had somehow been ditched, and Sol had been relegated to the mid-tier.
Now comes GPT-6.1 Sol, and the company says it’s more or less comparable with GPT-6.0 Astra in terms of its performance, especially in coding and intelligence-based tasks, the difference being that it’s much cheaper.
OpenAI presented a series of benchmark results to hammer home that point. For instance on DeepSWE 1.1, a benchmark to evaluate models’ software engineering performance, GPT-6.1 Sol achieved the same score as GPT-6 Astra at a cost of roughly one-fifth as many tokens per task. On the AutomationBench benchmark, which measures how well agents perform in multistep business workflows, GPT-6.1 Sol fell just short of GPT-6 Astra’s performance, but outperformed Anthropic PBC’s Claude Opus 5.5 while costing around one-third in terms of token consumption.
OpenAI also tested GPT-6.1 Sol on computer use tasks, where it beat out GPT-6 Sol by more than 7%, falling just short of GPT-6.1 Astra. The company said it also recorded fewer factual errors compared to GPT-6 Sol. The cost savings of using GPT-6.1 Sol instead of GPT-6.0 Astra are significant, OpenAI said. The new model is said to cost just 10 cent per million tokens, which is about half the price of GPT-6.0 Astra. It’s for this reason that OpenAI recommends the model for repeated and long-running workloads in coding and on application and document work.
Despite this, Astra remains OpenAI’s most powerful model, but the problem is that the company believes it’s so powerful, in fact, that it’s not safe to release it publicly at the moment. That’s why OpenAI announced yesterday it was delaying its release indefinitely, pointing to warnings from its own researchers and also third parties that have had the opportunity to study it that it had a tendency to do things without permission. It had planned to launch GPT-6.1 Astra today at its DevDay event, but the delay means that the new Sol gets all of the headlines instead.
The company said it’s making GPT-6.1 Sol available to all ChatGPT Work and Codex Plus, Pro, Business, Enterprise and Edu subscribers from today. It’s not yet available for the standard ChatGPT application, but should become an option there soon.
OpenAI also used its DevDay event to unveil a new set of “always-on” artificial intelligence agents called Dots, which feature cute, emoji-like avatars and serve their users as 24/7 digital assistants that can perform an expanded range of tasks.