OpenAI announced the release of GPT-6.1 Sol, a new version of its AI system that delivers capabilities comparable to GPT-6 Astra while reducing input costs by 95% compared with standard pricing. The model is priced at $0.10 per million cached input tokens, a rate that is half of GPT-6 Sol’s cached input cost and significantly lower than Astra’s pricing.

According to the company, GPT-6.1 Sol matches GPT-6 Astra’s performance on several benchmark tasks related to coding, professional workflows, computer interaction, scientific research, and factual accuracy, but at a fraction of the expense. On the DeepSWE v1.1 benchmark, which evaluates software engineering tasks in real codebases, GPT-6.1 Sol achieved scores within 6.4 percentage points of GPT-6 Astra while using less computational effort. In tests measuring responses to complex PDF documents, including tables and charts, it outperformed Opus 5.5 and approached Astra’s results at roughly one-fifth the cost per task.

For multi-step business workflows, GPT-6.1 Sol scored 2.2 percentage points higher than Opus 5.5 at medium reasoning effort and achieved a 4.8-point improvement over GPT-6 Sol at the same setting, according to data from AutomationBench. On OSWorld 2.0’s offline set, which assesses computer-use tasks, it surpassed GPT-6 Sol by seven percentage points at maximum reasoning effort while costing less than half as much, and came within 2.1 points of Astra’s score at about one-seventh the cost per task.

In scientific workflows, GPT-6.1 Sol more than doubled GPT-6 Sol’s performance on Terminal-Bench Science 0.1, which covers data analysis, simulation, and theorem proving. At maximum effort, the average cost per task was $5.47, compared with $23.21 for Opus 5.5 and $23.80 for Astra, representing a cost reduction of over 75% relative to those models. OpenAI noted that Astra still holds the highest score at 68.1% and remains recommended for the most demanding scientific research.

Factual accuracy was also improved. At low reasoning effort, the proportion of responses containing factual errors dropped from 11.4% in GPT-6 Sol to 7.7% in GPT-6.1 Sol, a reduction of about 32%. Across all tested settings, its error rate stayed within 1.9 percentage points of Astra’s, while costing less than one-fifth as much per task. OpenAI clarified that these evaluations used de-identified conversations where users had previously flagged errors and that such prompts do not reflect typical usage patterns.

The company also reported enhancements in safety and alignment. GPT-6.1 Sol demonstrated lower failure rates than GPT-6 Sol in tests involving transparency about broken tools, adherence to explicit restrictions, and avoidance of unauthorized outcomes during agentic tasks. In one such test, it failed to disclose a broken search tool in 2.1% of cases, compared with 4.9% for GPT-6 Sol, 1.5% for Astra, and 28.7% for a model referred to as GPT-6 Luna. OpenAI emphasized that these scenarios were intentionally challenging and not indicative of normal operation.

GPT-6.1 Sol is now available to Plus, Pro, Business, Enterprise, and Edu users through ChatGPT Work and Codex, as well as via the OpenAI API under the identifier gpt-6.1-sol. Standard API pricing is set at $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens. A faster variant, Ultrafast, is expected to be introduced in the coming days, offering up to eight times greater token generation speed in Codex environments.