GPT-6.1 Sol pushes the cost-efficiency frontier, per Artificial Analysis

Artificial Analysis, which scores models on its own Intelligence Index of evaluations, says OpenAI's GPT-6.1 Sol costs much less to run on that index than OpenAI's other recent models. According to its post, GPT-6.1 Sol at max effort costs $0.72 per Intelligence Index task. That is less than a quarter of GPT-6 Astra's $3.26, 31% below GPT-6 Sol's $1.04 and 64% below GPT-5.6 Sol's $1.99. The post followed the firm's earlier assessment of the model.

Cost here is quoted per Intelligence Index task, not per token, so it depends on how much work the model does on those tasks as well as on its prices. Effort is a setting that controls how much reasoning the model does before answering. Artificial Analysis says every effort level of GPT-6.1 Sol pushes out its cost-efficiency Pareto frontier, the line marking the cheapest option at each level of intelligence. In other words, at the score each setting reaches, there is no cheaper model. The post presents this as a frontier for OpenAI models, and the attached chart that would show the full comparison set was not collected.

The same post reports that OpenAI fixed a bug in how images were encoded for GPT-6 Luna and GPT-6 Sol, which means earlier results for those models were measured while the bug was in place. GPT-6 Luna at max effort gained one Intelligence Index point. The gains came on evaluations that benefit from visual understanding, including 4.1 points on MMMU-Pro and 2.4 points on GDP.pdf. GPT-6 Sol showed negligible improvement.