Google Now Has Two of the Top AI Models for Accounting Work, DualEntry Benchmark Finds

NEW YORK--(BUSINESS WIRE)--Google's newly released Gemini 4 Carbon scored 88.1% on the DualEntry Accounting AI Benchmark, tying for second among 24 leading AI models tested. Google's Gemini 4 Argon remains first at 91.1%.

Carbon shares second place with Zhipu AI's GLM-5.3 Flash and SpaceXAI's Grok 4.7. It finished ahead of Moonshot AI's Kimi K3.1 Pro at 86.1%, Anthropic's Claude Fable 5.1 at 85.1%, and OpenAI's best models, GPT-6.1 Sol and GPT-6 Astra, at 81.2%.

Google is the only provider with two models in the top four. A third Google model, Gemini 3.8 Flash, scored 83.2%.

Carbon's standout result is on the hardest work. It got 25 of 28 hard tasks right (89.3%), the best hard-task score of any model on the benchmark and ahead of Argon's 82.1%. Unusually, it scored slightly higher on hard tasks than on easy ones.

Key findings:

• Gemini 4 Carbon scored 88.1% overall, tied for second of 24
• It trails Gemini 4 Argon by 3.0 points overall, but beat it on hard tasks, 89.3% to 82.1%
• Its strongest area was transaction classification at 100%
• It matched Argon on journal entries at 92.3%, and scored 92.3% on financial reporting
• Its consistency score was 80.2%, behind Argon's 86.1%, the highest on the benchmark. It costs $0.081 per task, about 30% less than Argon ($0.115)
• It finished 5.9 points ahead of Claude Opus 5.5 (82.2%) and 6.9 points ahead of GPT-6 Astra (81.2%)
• Its weakest area was bank reconciliation at 66.7%, where Moonshot AI's Kimi K3.1 Pro scored 100%

Gemini 4 Carbon was competitive across most of the accounting cycle, scoring 83.3% in both accounts receivable and month-end close, the two hardest areas on the benchmark. It scored 84.6% in accounts payable and struggled most in bank reconciliation.

Its weakest result came in bank reconciliation at 66.7%, compared with 75% for Gemini 4 Argon and 100% for Kimi K3.1 Pro.

"Google now has the most accurate model we've tested and a second one close behind it," said Benedict Dohmen, co-founder of DualEntry. "For finance teams, the next gap that matters is consistency. Argon got the same answer in all three runs on 86% of tasks, and that is the bar an AI agent should clear before it touches the books."

The DualEntry Accounting AI Benchmark tests models on 101 synthetic accounting cases across eight areas: transaction classification, journal entries, accounts payable, accounts receivable, bank reconciliation, financial reporting, month-end close, and accounting knowledge.

Each model completes the full benchmark three times from an identical test company with the same tools and instructions. Answers are checked against reference answers written in advance using fixed scoring rules rather than an AI grader. No customer data is used.

Full results for all 24 models, including accuracy, consistency, cost, response time, and performance by accounting area, are available at dualentry.com/benchmark.

About DualEntry

DualEntry is the agentic, AI-native ERP for mid-market finance teams. It brings the general ledger, accounts payable and receivable, revenue recognition, multi-entity consolidation, and reporting into one platform, with AI agents that take on accounting work from data entry through close. Founded in 2024 and based in New York, DualEntry raised a $90 million Series A in 2025 from Lightspeed Venture Partners, Khosla Ventures, and GV. Learn more at DualEntry.com.