Condesion-Bench - Condition Satisfaction Rate: leaderboard

Condesion-Bench (2026) evaluates conditional decision-making in a compositional action space: for one trading day an LLM allocates quantities across 15 S&P 500 stocks under variable, contextual and allocation conditions, and its single zero-shot action is checked against the conditions. This board is the condition satisfaction rate in percent, the share of individual conditions the action satisfies; higher is better.

Metric: Condition Satisfaction Rate (%, from a 0-1 share times 100): share of the individual conditions satisfied, of the single action an LLM generates zero-shot (temperature 0, JSON output) on the 251 Condesion-Bench instances (one per trading day, October 2024 to September 2025: allocate a budget over 15 S&P 500 stocks under variable, contextual and allocation conditions); higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 16 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-5 Mini (2025-08-07)97.08#352
2GPT-596.41#185
3O3 (2025-04-16)96.28#238
4O4 Mini (2025-04-16)95.09#375
5GPT-OSS-20B88.58#1188
6GPT-OSS-120B87.92#712
7Claude 3.5 Sonnet (20241022)84.99#651
8Claude 3.5 Haiku82.34#1597
9GPT-4.1 Mini79.68#799
10GPT-4.179.68#506
11Mistral Large 2 (Jul)77.03#1164
12Llama 3.3 70B Instruct76.76#1394
13Gemini 2.0 Flash (001)64.67#733
14Mistral Small 3.262.02#1647
15Gemini 2.0 Flash Lite (001)61.75#1122

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

Interactive version: theaggregate.ai/benchmark?slug=condesion-bench-condition-satisfaction-rate · How It Works · Data refreshed daily, snapshot 2026-11-22.