OpenAI releases GPT-6.2 Astra and Sol with lower prices and modest benchmark gains
OpenAI has released two new models, GPT-6.2 Astra and Sol, which are more capable and cheaper to use than their predecessors. Astra ranks second in an independent intelligence index, and OpenAI itself flags it as a risk.
OpenAI has released two new models, GPT-6.2 Astra and GPT-6.2 Sol, which the company says are both more capable and cheaper to use than the models that came before them. The launch lowers the price of access through OpenAI's application programming interface, the channel developers and businesses use to build their own products on top of the company's AI systems, and it brings measurable gains on several benchmark tests.
The most notable result comes from the independent Artificial Analysis Intelligence Index, where Astra ranks second, behind Claude Fable 5.2. Astra is also described as the first publicly released model to reach the High threshold for AI self-improvement on OpenAI's own internal scale. That is a classification OpenAI itself treats as a risk, meaning the company has acknowledged that a model at this level warrants closer attention than its predecessors, even as it puts the model into commercial release.
The pricing change matters as much as the benchmark results for companies deciding which models to build on. Lower API costs make it cheaper for startups and established firms to run the models at scale, which tends to widen the range of applications that become economically viable. The two new models are positioned as successors to earlier GPT generations, with Astra aimed at the strongest performance and Sol at a combination of capability and cost.
The release also sits within a wider competitive race among the leading AI developers, where companies routinely announce new models with improved scores on standardized tests. Rankings on independent indexes give buyers a comparison point that the companies' own marketing cannot provide, which is why the placement of Astra behind a rival model is relevant to the market. The gains reported are described as some benchmark improvements rather than across-the-board leaps, and the company's own risk classification adds a note of caution to the announcement.
For users, the immediate effect is access to newer models at lower cost. For the broader industry and policymakers, the self-improvement classification is likely to draw the most attention, since it points to capabilities that developers are now tracking explicitly as potential hazards. The launch of GPT-6.2 Astra and Sol therefore combines a commercial step forward in pricing and performance with a public acknowledgment of the risks that the most advanced systems carry.
Source: the-decoder.com
3
