FINESSE-Bench - VLigaBench-ru: leaderboard

FINESSE-Bench VLigaBench-ru tests financial knowledge with 324 Russian-language olympiad problems in micro- and macroeconomics, financial mathematics and game theory (numerical and short answers) from FINESSE-Bench (2026). Answers are marked correct or incorrect by a GPT-5.2 judge against the reference; the score is accuracy in percent, higher is better.

Metric: Accuracy (%) on the 324 VLigaBench-ru Russian-language olympiad problems in micro- and macroeconomics, financial mathematics and game theory (numerical and short answers) of FINESSE-Bench; a GPT-5.2 judge marks each answer correct or incorrect against the reference; zero-shot, one fixed prompt per task type, temperature 0 where possible, reasoning configurations with medium effort where the model offers them; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 31 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Qwen 3.5 397B A17B84.88#264
2GPT-5.2Medium84.88#185
3Qwen 3.5 Plus (2026-02-15)84.88#194
4Qwen 3.5 122B A10B83.64#346
5Kimi K2.583.02#266
6GLM-4.782.41#380
7GLM-582.1#269
8Qwen 3.5 27B81.17#375
9Qwen 3.5 Flash (02-23)81.17#557
10Qwen 3.5 35B A3B80.56#492
11Claude Sonnet 4.6Medium80.56#173
12Qwen 3 235B A22B 2507 (Thinking)79.32#466
13GPT-5.4Medium79.32#135
14MiniMax-M2.579.01#600
15Claude 3.7 SonnetThinking79.01#470

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

Interactive version: theaggregate.ai/benchmark?slug=finesse-bench-vligabench-ru · How It Works · Data refreshed daily, snapshot 2026-11-21.