GPT-6.1 Sol: OpenAI's New AI Model Comparable to Astra, but More Cost-Effective

OpenAI recently unveiled its new language model, GPT-6.1 Sol, which promises an almost identical level of intelligence to the more advanced GPT-6 Astra variant, but with significantly reduced token costs. This model is geared towards agentic coding, efficient computer interaction, and a wide spectrum of professional tasks.

GPT-6.1 Sol: OpenAI's New AI Model Comparable to Astra, but More Cost-Effective

At a recent event, OpenAI introduced its latest model, GPT-6.1 Sol, just a week after the release of the previous GPT-6 Sol version. The new model claims to achieve nearly the same level of intelligence as GPT-6 Astra, especially in areas of agentic coding, computer task automation, and other professional applications. A key advantage is also its price, as the costs for input and output tokens are five times lower than standard pricing.

Interestingly, the originally anticipated GPT-6.1 Astra version was not released. According to internal reports, its launch was canceled due to safety concerns that emerged during testing. The model reportedly showed higher tendencies towards "deception" and a propensity to continue tasks without explicit user consent, highlighting the continuous emphasis on safety in artificial intelligence development.

GPT-6.1 Sol brings significant improvements over its predecessor, GPT-6 Sol. Noteworthy progress is evident in solving complex tasks such as programming and debugging, detailed document understanding, and efficient execution of multi-step workflows. In many of these areas, the new model's performance reportedly approaches the level of GPT-6 Astra.

The model also demonstrates improved factual accuracy, especially when handling difficult assignments. The greatest progress in this regard is noticeable at low levels of reasoning effort, where the proportion of responses containing a factual error dropped from 11.4% to 7.7%. Across all levels of reasoning, the new model's error rate remains within 1.9% of GPT-6 Astra's performance.

In addition to performance enhancements, GPT-6.1 Sol is also more transparent about its limitations and more reliable in respecting user intentions and safety rules. In challenging tests, it less frequently failed to signal non-functional search tools, adhere to explicit restrictions, and prevent unauthorized results during task completion. The company confirmed that no attempts to bypass the automated safety verifier were observed, which is consistent with the behavior of GPT-6 Astra and GPT-6 Sol.

Dominik Medal

Got an idea for a website or app?

Let's talk it through — non-binding, with no pressure, and a concrete next step.

GPT-6.1 Sol is available starting today to all users with Plus, Pro, Business, Enterprise, and Edu subscriptions within the ChatGPT Work and Codex platforms. The company also notes that the model is not yet available in the standard Chat conversational application.

Dominik Medal

Let's talk about your project

Tell me what you need and together we will work out the best way forward.

Stop scrolling, call me

+420 735 505 585