JFinBench - No Tools - CPA: leaderboard

321examples, 25sections, 6examfamilies; mean3runs. OneAPIcall/question, 131072outputtokens(65536forGemini), nativeeffort/thinking retained. MCQ extraction; GPT5.1 judging freeanswers undernativepartialcreditrubrics; exammean overexamples, overallmeanover6exams. Failure attempts0 pernativepolicy. Public321taskset reconstructablefromofficialsourcePDFrecipes; HFaloneomits57examples/55answers. Metric: Score (%). Source: https://arxiv.org/html/2611.18130v1

Metric: Score (%). Source: arxiv.org. Saturation forecast: Around February 2027. 12 models tracked. Updated 17 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Fable 5Max97.1#28
2GPT-6 AstraMax96.8#8
3Gemini 3.8 FlashHigh92.5#39
4Kimi K3.1 ProMax84.2#32
5Kimi K3Max78.9#59
6DeepSeek V4.1 Flash (1017)Max73.1#44
7Qwen 3.8 FlashThinking65.9#128
8MiMo-V2.6-ProThinking63.4#86
9GLM-5.3 FlashMax56.6#136
10Mistral Large 4Thinking54.8#143
11Kimi K2.6Thinking54.5#191
12MiMo-V2.6-FlashThinking50.5#176

No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=jfinbench-no-tools-cpa · How It Works · Data refreshed daily, snapshot 2026-11-17.