GPT-6.1 Terra Is Fast and Cheap. On Accounting, It's Not Accurate Enough to Lead

OpenAI just launched GPT-6.1 Terra, and we ran it through the DualEntry Accounting AI Benchmark, which tests leading AI models on 101 accounting tasks across eight areas including journal entries, accounts payable, accounts receivable, bank reconciliation, financial reporting, and month-end close.

‍

GPT-6.1 Terra scored 78.2% accuracy, tying for 18th of 22 models. That puts it behind OpenAI's own GPT-6.1 Sol and GPT-6 Astra (both 81.2%), and well behind the leader, Google's Gemini 4 Argon, at 91.1%. It finished ahead of Claude Haiku 5.5 and GPT-6 Sol (both 76.2%) and GPT-6 Luna (66.3%).

‍

Accuracy isn't where Terra stands out. Speed and cost are. It's the fastest OpenAI model we've tested and the third fastest overall, at a fraction of what OpenAI's larger models cost per task.

Where GPT-6.1 Terra performed best

‍

GPT-6.1 Terra's strongest results came in:

‍

• Transaction classification: 100% (13 of 13)
• Journal entries: 92.3% (12 of 13)
• Bank reconciliation: 83.3% (10 of 12)

‍

The reconciliation result is the surprise. Only Kimi K3.1 Pro (100%) and Muse Wonder (91.7%) scored higher, and Terra matched Grok 4.7 and Claude Fable 5.1 there.

‍

It also did better on harder work than its overall rank suggests. Terra got 31 of 36 medium tasks right (86.1%) and 20 of 28 hard tasks (71.4%), with only eight and seven models doing better on each.

Fast, and fairly cheap

‍

GPT-6.1 Terra averaged 16.1 seconds per task, the third fastest of any model we've tested, behind Claude Haiku 5.5 (11.4s) and Claude Sonnet 5.5 (12.1s). That's roughly half the time of GPT-6.1 Sol (27.2s) and GPT-6 Astra (34.3s).

‍

It costs $0.042 per task, or $12.73 for a full three-run benchmark. That's the seventh cheapest result, about a tenth of GPT-6 Astra ($0.49) and half of GPT-6.1 Sol ($0.09).

‍

It isn't the best value, though. DeepSeek V4.1 Flash, GLM-5.3 Flash and GLM-5.3 all cost less per task and score higher, and Mistral Large 4 costs about the same while scoring two points higher.

Where GPT-6.1 Terra struggled

The biggest weaknesses came in:

‍

• Accounts payable: 53.8% (7 of 13), tied for last with GPT-6 Luna
• Accounts receivable: 58.3% (7 of 12)
• Month-end close: 58.3% (7 of 12)
• Financial reporting (76.9%, 10 of 13) was also below average.

‍

The more unusual weakness was easy tasks. Terra got 28 of 37 right (75.7%), and only GPT-6 Luna did worse. It lost most of its points on simple cases, not hard ones.

‍

Those misses pulled its overall score down to 78.2%.

How consistent was GPT-6.1 Terra?

GPT-6.1 Terra passed 65.3% of tasks in all three benchmark runs, the second lowest result after GPT-6 Luna.

‍

For comparison:

‍

• GPT-6.1 Sol: 68.3%
• GPT-6 Astra: 71.3%
• Gemini 4 Argon (leader): 86.1%

‍

That matters because Terra may get a task right once, but a low consistency score means it's less likely to get the same work right every time. For accounting, where an agent may run the same task many times a month, that gap shows up as review work.

Where GPT-6.1 Terra sits now

GPT-6.1 Terra is currently tied for 18th of 22 on the DualEntry Accounting AI Benchmark at 78.2% accuracy.

‍

It's competitive in transaction classification, journal entries and bank reconciliation, and it's one of the fastest models we've tested. But weak results in payables, receivables and month-end close, plus misses on easy tasks, keep it well behind the leaders.

‍

For teams that need fast, low-cost answers on structured tasks like classification, Terra is worth a look. For close and receivables work, where getting it right every run matters, more accurate and more consistent models are available at a similar price.

‍

See the full DualEntry Accounting AI Benchmark to compare every model across accuracy, consistency, cost, speed, and accounting task area.