Google's Gemini 4 Argon got 91.1% of accounting tasks right in the first full round of the DualEntry Accounting AI Benchmark, the highest score among 18 leading AI models. It was also the most consistent, passing 86.1% of tasks in all three test runs.

Zhipu AI's GLM-5.3 Flash and SpaceXAI's Grok 4.7 tied for second at 88.1%. Anthropic's best model, Claude Fable 5.1, scored 85.1%. OpenAI's best, GPT-6 Astra and GPT-6.1 Sol, scored 81.2%.

Key findings:

• Gemini 4 Argon had the highest accuracy (91.1%) and the highest consistency (86.1%)
• Argon led or tied in five of eight accounting areas, with 100% on transaction classification, financial reporting and accounting knowledge
• On hard tasks, Argon scored 82.1%, tied with Grok 4.7 and Gemini 3.8 Flash for the best result
• Accounts receivable and month-end close were the hardest areas, where models averaged 64.4%. Receivables scores ranged from 41.7% to 91.7%
• Claude Fable 5.1 cost $1.18 per task, about 10 times Argon's $0.11, for a lower score
• DeepSeek V4.1 Flash scored 84.2% at $0.003 per task, about 400 times cheaper than Claude Fable 5.1

Argon was not the best at everything. Its weakest area was bank reconciliation, at 75.0%, where Meta's Muse Wonder led at 91.7%. Moonshot AI's Kimi K3 led month-end close at 91.7%, ahead of Argon's 83.3%.

Consistency separated models with similar headline scores. Grok 4.7 matched GLM-5.3 Flash on accuracy but passed only 74.3% of tasks in all three runs.

"A model that gets a journal entry right once and wrong the next time isn't ready for the books," said Benedict Dohmen, co-founder of DualEntry. "Argon was the most accurate model we tested and the most consistent, and finance teams need both."

The benchmark runs on DualBench100, DualEntry's published evaluation method. Each model completes 101 synthetic accounting cases across eight areas, three times, starting each run from an identical test company with the same tools and instructions. Answers are checked against reference answers written in advance, using fixed rules rather than an AI grader. A task counts as passed when the model passes it in at least two of three runs. No customer data is used.

Full results for all 18 models, including cost, consistency and scores by task area, are 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.

View source version on businesswire.com: https://www.businesswire.com/news/home/20261014943739/en/