GPT-6.1 Terra trails leading models on accounting work in independent DualEntry benchmark

#What happened

OpenAI's newly released GPT-6.1 Terra scored 78.2% on the DualEntry Accounting AI Benchmark, tying for 18th among 22 leading AI models tested. Its strongest area was transaction classification at 100% (13 of 13), while its weakest was accounts payable at 53.8% (7 of 13), tied for last. It trailed Google's Gemini 4 Argon at 91.1%, Zhipu AI's GLM-5.3 Flash at 88.1%, SpaceXAI's Grok 4.7 at 88.1%, and OpenAI's own GPT-6.1 Sol at 81.2%. It scored 71.4% on hard tasks but only 75.7% on easy ones, with a consistency score of 65.3%.

#Why it matters

Terra's standout is speed and cost, not accounting accuracy. At 16.1 seconds per task it was the fastest OpenAI model on the benchmark and the third fastest overall, at $0.042 per task, about half of GPT-6.1 Sol and a tenth of GPT-6 Astra. For finance work where accuracy matters, this result argues for checking Terra's output rather than trusting it outright.

Verified October 26, 2026.