OpenAI has officially released GPT-6.1 Sol, an advanced iteration of its GPT-6 Sol model, promising a significant leap in performance while drastically cutting operational costs. This new model is engineered to deliver capabilities approaching those of the more powerful GPT-6 Astra, but at approximately one-fifth of the price, making sophisticated AI more accessible for a broader range of applications.
Targeting Complex Workflows with Enhanced Efficiency
GPT-6.1 Sol is not merely an incremental update; it's designed to tackle demanding tasks far beyond basic conversational AI. Its primary focus areas include complex coding challenges, intricate business workflows, computer-based automation, and rigorous scientific research. This strategic emphasis positions GPT-6.1 Sol as a powerful tool for developers building sophisticated AI agents that require multi-step reasoning and execution.
According to OpenAI, GPT-6.1 Sol can match Astra's performance in several key metrics. For instance, in coding assessments like DeepSWE v1.1, it achieves Astra-level results at a fraction of the cost, significantly outperforming its predecessor, GPT-6 Sol, by 6.4 percentage points. Similarly, for document-based tasks (GDP.pdf), it surpasses Opus 5.5 with fallbacks at less than half the cost, nearing Astra's efficacy.
Improved Accuracy and Safety Protocols
Beyond raw performance, GPT-6.1 Sol also introduces notable improvements in factual accuracy, transparency, and safety. OpenAI reports a reduction in factual errors from 11.4% to 7.7% at low reasoning efforts, based on deliberately challenging prompts. The model also demonstrates better handling of user intent and adheres more strictly to safety restrictions, showing no attempts to bypass automated safety reviews.
Availability and Cost for Developers
GPT-6.1 Sol is currently available to Plus, Pro, Business, Enterprise, and Edu users within ChatGPT Work and Codex platforms. Developers can access the model via API at a competitive price point: $2 per million input tokens, $0.10 for cached input, and $10 for output tokens. The cached-input pricing represents a 95% reduction from standard input costs and is 50% cheaper than GPT-6 Sol's cached-input rate, offering substantial savings for repetitive tasks.
OpenAI also announced an Ultrafast version of GPT-6.1 Sol, which will be integrated into Codex. This variant is expected to offer up to eight times faster token generation compared to the standard speed, further enhancing its utility for time-sensitive applications and high-throughput environments.