ChatGPT 6.1 Sol Lands With Astra Level Skill at One Fifth the Price

On a Tuesday, OpenAI revealed a new model for its paying ChatGPT users. ChatGPT 6.1 Sol appears to be tracking rather well in several areas, such as coding jobs, dense document reading, as well as long stretches of computer work, and all at a token cost that is around one-fifth of what Astra, the other model, charges. The release occurred during the DevDay event in San Francisco, only one day after OpenAI revealed that they had been holding back a 6.1 version of Astra because to safety concerns, including the danger that it would do things without your permission.

OpenAI subscribers, which include Pro, Business, Enterprise, and Edu users, now have access to Sol within the tools they already use to get work done, such as ChatGPT Work and Codex. That’s a huge benefit for customers because it allows them to get on with their lives without having to deal with a second app. They’re already familiar with the tools they use at work, so this makes things easier. Sol performed admirably when it came to difficult software tasks run through actual codebases, finishing just as high as Astra and outscoring its predecessor by 6.4 points. Sol held its own with Astra and even outperformed Anthropic’s Opus 5.5 when it came to extracting answers from very dense PDFs packed with tables, charts, and diagrams, the type of complexity found in finance, healthcare, or law, all for roughly half the price that would have been charged for that test.

When it comes to more complex business tasks that require a variety of tools, such as sales, marketing, operations, and support, Sol really impressed, gaining 4.8 points over its predecessor and beating Opus 5.5 by 2.2 points, all at a fraction of the previous Sol’s price. When it comes to computer-use runs, when the model has to go through a software program and perform a multi-step task, Sol made a significant jump ahead, closing in on Astra by 2.1 points at a cost that was one-seventh of what Astra charged.
[Source]