OpenAI Launches GPT-6.1 Sol

OpenAI has introduced GPT-6.1 Sol, which offers a 95% reduction in standard input pricing compared to previous models, positioning it as a cost-effective alternative for everyday tasks. The cached input cost is set at $0.10 per million tokens, which is 50% lower than GPT-6 Sol.

Performance-wise, GPT-6.1 Sol shows a 6.4 percentage point improvement in the DeepSWE v1.1 evaluation tool for software-engineering tasks and a 4.8 percentage point increase in AutomationBench scores over GPT-6 Sol. Additionally, it achieves a 7 percentage point improvement in OSWorld 2.0 evaluations.

The model also demonstrates a significant reduction in factual errors, decreasing from 11.4% to 7.7%, marking a 32% reduction. Its error rate is now within 1.9 percentage points of GPT-6 Astra, which is notable given that GPT-6.1 Sol is priced at roughly one-fifth of the cost of GPT-6 Astra, which costs $23.80 per task.

For specific tasks, GPT-6.1 Sol has a cost of $5.47 per task in Terminal-Bench Science 0.1, compared to $23.21 for Opus 5.5 and $23.80 for GPT-6 Astra. OpenAI claims that GPT-6.1 Sol scores higher than Opus 5.5 while maintaining a cost per task of less than half.

The model is also capable of ultrafast token generation, achieving speeds up to 8 times faster than standard rates. This launch follows the recent cancellations of the GPT-6 Astra release due to safety concerns, emphasizing OpenAI's focus on improving both capability and cost-effectiveness in its AI offerings.