GPT-6.1 Terra is significantly better on the Agents' Last Exam, completing tasks cheaper and faster

The GPT-6.1 Terra model significantly outperforms GPT-5.6 Terra on the Agents' Last Exam benchmark, which measures agents' software-usage ability. GPT-6.1 Terra achieves the same score as GPT-6 Astra but is ten times cheaper per token and completes tasks on average more than three times faster, providing a cost and efficiency advantage.