Meituan LongCat Team Unveils General 365 Reasoning Benchmark as Most AI Models Struggle to Pass
The Meituan LongCat team has officially introduced General 365, a rigorous new benchmark designed to measure the reasoning capabilities of modern artificial intelligence. In an initial evaluation of 26 prominent models, the benchmark has proven to be a significant challenge for the industry. Even the high-performing Gemini 3 Pro, currently regarded as one of the strongest models available, only managed an accuracy rate of 62.8%. Alarmingly, the vast majority of the other tested models failed to reach a 60% score, which serves as the passing threshold for this evaluation. This release marks a pivotal shift in how AI reasoning is measured, suggesting that current mainstream models still have substantial progress to make in complex logical processing and highlighting a significant gap in the industry's current reasoning capabilities.
Key Takeaways
- New Evaluation Standard: Meituan's LongCat team has launched General 365, a benchmark specifically designed to set a new standard for AI reasoning.
- Industry-Wide Struggle: Out of 26 mainstream models tested, the majority failed to achieve a passing score of 60%.
- Leading Performance: Gemini 3 Pro emerged as the top performer in the test, yet it only achieved an accuracy rate of 62.8%.
- Reasoning Gap: The results highlight a significant disparity between current AI capabilities and the requirements of advanced reasoning tasks.
In-Depth Analysis
The Rigor of the General 365 Benchmark
The introduction of General 365 by the Meituan LongCat team represents a significant development in the field of artificial intelligence evaluation. By positioning this benchmark as a "new ruler" for reasoning, the team has established a high bar that most current models are unable to meet. The core finding of their initial report is that the vast majority of the 26 mainstream models tested could not reach the 60% accuracy mark. This 60% threshold, often considered a basic passing grade in many academic and professional contexts, appears to be a major hurdle for contemporary large language models when applied to the specific reasoning challenges presented by General 365.
The fact that so many models failed to reach this baseline suggests that General 365 targets specific logical and reasoning complexities that are not fully addressed by existing training methodologies or datasets. It serves as a stark reminder that while AI models are becoming increasingly proficient at language generation and pattern recognition, their ability to perform consistent, high-level reasoning remains a work in progress.
Benchmarking the Leaders: The Case of Gemini 3 Pro
One of the most notable aspects of the General 365 release is the performance data regarding Gemini 3 Pro. Identified in the report as the "strongest model on the planet" currently, Gemini 3 Pro's performance serves as a ceiling for the current state of the art. However, even this leading model only achieved an accuracy rate of 62.8%.
This specific data point is crucial for understanding the current landscape of AI. If the industry leader is only barely clearing the 60% passing line, it indicates that the "reasoning ceiling" for current AI technology is lower than many might have anticipated. The 62.8% score of Gemini 3 Pro, while superior to its peers, emphasizes that even the most advanced systems have significant room for improvement. The gap between 62.8% and a near-perfect score represents the remaining frontier for AI development in the realm of complex reasoning and logical deduction.
Industry Impact
The release of General 365 is likely to have a profound impact on how AI developers and researchers approach the problem of reasoning. By providing a benchmark where even the most advanced models struggle, Meituan's LongCat team has created a tool that can more effectively differentiate between models that merely simulate reasoning and those that possess genuine logical processing capabilities.
For the AI industry, these results serve as a call to action. The widespread failure to meet the 60% mark suggests that current evaluation metrics may be overestimating the actual reasoning proficiency of mainstream models. General 365 provides a more grounded and challenging metric that will likely drive future research toward improving the underlying cognitive architectures of AI. As teams strive to move their models past the 60% threshold and closer to the performance of Gemini 3 Pro—and eventually beyond—the industry will see a renewed focus on the quality of reasoning over the quantity of parameters.
Frequently Asked Questions
Question: What is the General 365 benchmark?
General 365 is a new reasoning evaluation benchmark released by the Meituan LongCat team. It is designed to measure the logical reasoning capabilities of large language models and has established itself as a rigorous "new ruler" for the industry.
Question: How did mainstream AI models perform on this benchmark?
In a test of 26 mainstream models, the performance was generally low. Most models failed to reach a 60% accuracy rate. The top-performing model, Gemini 3 Pro, achieved a score of 62.8%, highlighting the difficulty of the benchmark.
Question: Who developed General 365?
General 365 was developed and released by the Meituan LongCat team, part of Meituan's technical division.
