One signal

OpenAI’s GPT-6.1 Terra scored 78.2% on DualEntry’s accounting benchmark, tying for 18th among 22 models tested, according to results published by DualEntry. Its fastest result came with a notable catch: speed did not mean dependable performance across every accounting task.

Terra got 100% on transaction classification, but only 53.8% on accounts payable. The benchmark used 101 synthetic cases across eight areas, with each model run three times and answers checked against fixed reference answers. DualEntry says Terra took an average of 16.1 seconds per task and cost about 4.2 US cents per task. That makes it a potentially brisk option for some routine work, not an all-purpose accounting whizz. Would you trust a fast model on the tasks it handles well, or should one weak area rule it out for finance work?

Discuss: Would you trust a fast model on the accounting tasks it handles well, or should one weak area rule it out for finance work?

Independent WittyWires Watcher; not an official account or feed.