Google Is Dominating Accounting AI. Three Gemini Models Now Rank Among the Top Five

John Iwuozor leads content, finance, and research at DualEntry. He writes for the people who actually close the books: controllers running month-end, CFOs weighing an ERP migration, accountants buried in reconciliations. Before DualEntry he spent five years covering B2B SaaS and fintech for Forbes Advisor, Ramp, and Infosecurity Magazine. He builds every piece on primary data, including benchmarks, practitioner interviews, and product testing, rather than recycled advice. He is also an avid chess player.

Google is tightening its grip on the top of the DualEntry Accounting AI Benchmark. Its latest model, Gemini 4 Neon, scored 91.1% across 101 accounting tasks, tying Gemini 4 Argon for first place.
That gives Google three models among the five highest-scoring performers we've tested, with Gemini 4 Carbon also sitting at 88.1%.

The gap with other major AI providers is becoming harder to ignore. Even Carbon outperformed Anthropic's Claude Fable 5.1 (85.1%) and OpenAI's GPT-6.2 Sol (84.2%).
Neon also ranked second on consistency and hard accounting tasks, although month-end close remains a weakness.
Key findings
- Gemini 4 Neon scored 91.1% accuracy, tying Gemini 4 Argon for the highest score on our accounting benchmark.
- Three Google models now rank among the top five, with Gemini 4 Carbon scoring 88.1%.
- Neon achieved perfect scores in transaction classification, financial reporting, and accounting knowledge.
- It ranked second on hard tasks at 85.7% and consistency at 85.1%, behind Carbon and Argon, respectively.
- Month-end close was Neon's weakest area at 66.7%, compared with Argon's 83.3%.
Where Gemini 4 Neon performed best
Neon passed 92 of the benchmark's 101 accounting tasks, matching Argon's overall result.
It achieved perfect scores in three of the eight accounting areas:
- Transaction classification: 100% (13/13)
- Financial reporting: 100% (13/13)
- Accounting knowledge: 100% (13/13)
It also performed well in journal entries (92.3%), accounts receivable (91.7%), and bank reconciliation (91.7%).
Month-end close remains Neon's biggest weakness
Despite its first-place finish, Neon struggled with month-end close, completing only eight of the twelve tasks for 66.7% accuracy.
That puts it behind Argon and Carbon, which both scored 83.3%, and Kimi K3, which currently leads the category at 91.7%.
Month-end close was also Neon's only accounting category below 80%.
The result reinforces a pattern we've seen across the benchmark: the highest-scoring models don't necessarily lead every accounting workflow.
What about cost and speed?
Neon's strong accuracy comes at a relatively high cost.
The model cost $52.76 across three complete benchmark runs, averaging approximately $0.17 per task.

It averaged 24.3 seconds per task, ranking 12th fastest among the 27 models compared.
Those figures matter when evaluating models for accounting workflows that may involve hundreds or thousands of repeated actions.
Is Google pulling away in accounting AI?
Google currently has the strongest lineup on our accounting benchmark, but it hasn't swept every category.
Kimi K3.1 Pro leads bank reconciliation with 100%, Kimi K3 leads month-end close at 91.7%, and Gemini 3.8 Flash remains the only model with a perfect score on journal entries.
The benchmark uses 101 synthetic accounting cases, so these rankings reflect performance under specific testing conditions rather than every possible real-world accounting workflow.
Still, the latest results show just how far ahead Google has pulled in overall accounting accuracy, even as competitors continue to outperform its models on individual tasks.
Explore the full DualEntry Accounting AI Benchmark to compare the latest models across accuracy, consistency, accounting tasks, speed, and cost.



.jpg)