NEW YORK: Meta's new Muse Wonder model got 83.2% of tasks right on the DualEntry Accounting AI Benchmark, tying for second place among 57 AI models tested. Its predecessor, Muse Spark 1.1, scored 61.4% on the same tasks.

Meta released Muse Wonder on October 12 and called it its most capable model to date. On accounting work, it now sits ahead of every OpenAI and Google model on the benchmark and level with Anthropic's Claude Fable 5. Only Grok 4.5 scored higher, at 84.2%.

Key results:

• Overall accuracy: 83.2%, up from 61.4% for Muse Spark 1.1
• Financial reporting: 97%
• Bank reconciliation: 92%
• Easy, medium and hard tasks: 86%, 87% and 77%
• Best OpenAI model on the benchmark: GPT-5.5 Pro at 80.2%
• Best Google model: Gemini 3.7 Flash at 82.2%

Muse Wonder was most reliable on reporting. It answered almost every reporting question correctly across repeated runs, including cash flow and financial statement calculations.

Its misses clustered in day-to-day transaction work. It got some accounts receivable payment scenarios wrong and was inconsistent on month-end close journal entries and edge cases in transaction classification. A strong overall score can hide those gaps, and they are the details a finance team's books depend on.

"Muse Wonder reads financial statements as well as any model we've tested," said Benedict Dohmen, co-founder of DualEntry. "Where it still slips is the receivables and journal entries underneath them, which is exactly the work finance teams want an agent to take off their plate."

The DualEntry Accounting AI Benchmark tests models on 101 synthetic accounting cases across eight areas: transaction classification, journal entries, financial reporting, bank reconciliation, accounts payable, accounts receivable, month-end close and accounting knowledge. The cases split into 37 easy, 36 medium and 28 hard. Each case has a fixed reference answer and is scored by deterministic rules with an 80% pass mark, and every model runs the full set several times to measure consistency. The methodology and results for all 57 models are at dualentry.com/accounting-ai-benchmark.

Fonte: Business Wire