Anthropic’s New Claude Fable 5.2 Is Worse At Accounting Than The Model It Replaced
Anthropic‘s newest model, Claude Fable 5.2, scored 81.2% on accounting firm DualEntry‘s AI benchmark, nearly 4 points worse than the Claude Fable 5.1 model it replaced, which scored 85.1%.
Key Takeaways
- Claude Fable 5.2 scored 81.2% on DualEntry’s Accounting AI Benchmark, against 85.1% for Claude Fable 5.1
- That ties Fable 5.2 for 15th place among the 24 models on the leaderboard, while Fable 5.1 sits 6th
- Google’s Gemini 4 Argon leads the ranking with 91.1%
- Fable 5.2 still got every transaction classification task right
DualEntry, which makes accounting software, published the result on its public leaderboard, which tests AI models on 101 real-world style accounting tasks.
How Anthropic’s Models Rank
Fable 5.2 shares 15th place with OpenAI’s GPT-6.1 Sol and GPT-6 Astra, which also scored 81.2%. Grok 4.7, GLM-5.3 Flash and Gemini 4 Carbon tie for second at 88.1%.
Also Read: Anthropic Cuts Claude Fable Price 20%, Says It Can Match Opus For A Tenth Of The Cost
What The Test Measures
The 101 tasks cover eight areas, including journal entries, bank reconciliation, accounts payable and receivable, month-end close and financial reporting. Answers are scored against fixed reference answers, and a model passes a task only if it gets it right in at least two of three runs.
The ranking covers one company’s accounting test only. It does not measure how Fable 5.2 handles coding, writing or other work, and DualEntry sells the kind of accounting automation the benchmark grades.
Read Next: Anthropic Chases $64B In Second-Biggest IPO Ever, Behind Only SpaceX
