OpenAI has introduced GPT-6.1 Sol, an upgrade to the existing GPT-6 Sol model that aims to bridge the gap between cost and high-end performance. The company reports that the new model nearly matches the intelligence of GPT-6 Astra on agentic coding, computer use, and professional work tasks, but at approximately one-fifth of Astra’s standard input and output token prices.

What Happened

GPT-6.1 Sol is designed to offer a more balanced ratio of capability to cost for everyday professional and development tasks. According to OpenAI, the model delivers substantial improvements over GPT-6 Sol in complex areas such as writing, debugging code, and executing multi-step business workflows. On the DeepSWE v1.1 benchmark, which evaluates software engineering tasks in real codebases, GPT-6.1 Sol reportedly matches GPT-6 Astra’s performance at roughly one-fifth of the cost, while exceeding GPT-6 Sol’s best score by 6.4 percentage points.

In evaluations focused on professional document analysis, specifically the GDP.pdf benchmark, GPT-6.1 Sol scores higher than Opus 5.5 with fallbacks at less than half the cost per task. The model also approaches GPT-6 Astra’s state-of-the-art performance in this area at roughly one-fifth the cost. For automation tasks measured by AutomationBench, the company states that GPT-6.1 Sol scores 2.2 percentage points above Opus 5.5 at medium reasoning effort, at about a third of the cost, and shows a 4.8 percentage point improvement over GPT-6 Sol.

The update also includes significant gains in computer use and scientific research capabilities. On OSWorld 2.0’s offline set, GPT-6.1 Sol outperforms GPT-6 Sol by seven percentage points at maximum reasoning effort, coming within 2.1 percentage points of Astra’s score at roughly one-seventh the cost per task. In scientific workflows evaluated by Terminal-Bench Science 0.1, the model more than doubles GPT-6 Sol’s score at maximum reasoning effort, costing an average of $5.47 per task compared to $23.21 for Opus 5.5 and $23.80 for Astra.

OpenAI also highlights improvements in factual accuracy. At low reasoning effort, the company reports that GPT-6.1 Sol reduces the share of responses containing factual errors from 11.4% to 7.7% compared to GPT-6 Sol. Across tested settings, its error rate remains within 1.9 percentage points of GPT-6 Astra’s, at less than one-fifth the cost per task.

Why It Matters

For developers and enterprises, the pricing structure of GPT-6.1 Sol presents a potential shift in the economics of deploying AI agents. With standard API prices set at $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens, the model offers a lower barrier to entry for high-performance agentic applications. The cached input cost is notably 95% less than standard input pricing, which OpenAI suggests gives developers more room to build agents that reuse context across requests.

The model’s availability starts today for Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex, though it is not yet available in standard Chat. Developers can access the model via the OpenAI API as gpt-6.1-sol. OpenAI also announced that an Ultrafast version, offering up to 8x faster token generation, will be available in the coming days.

From a safety perspective, OpenAI reports that GPT-6.1 Sol shows improvements in alignment evaluations, becoming more transparent about its limitations and more reliable in respecting user intent. The company states that the model shows lower failure rates than GPT-6 Sol in challenging evaluations regarding broken search tools and unauthorized outcomes during agentic tasks, with no observed attempts to bypass automated safety reviewers.

The Bottom Line

GPT-6.1 Sol positions itself as a cost-effective alternative to GPT-6 Astra for many professional and agentic tasks, though OpenAI notes that GPT-6 Astra still achieves the highest scores in the most difficult scientific research evaluations. The release emphasizes a strategic balance of performance and price, aiming to make high-level AI capabilities more accessible for widespread developer and enterprise use.