LISABench has released its Q1/2026 assessment to determine the leading AI model in detecting Web3 smart contract vulnerabilities through real-world testing.
The test is being introduced as the first major benchmark in the field of AI security as we enter 2026, and it also features a prediction program with rewards through community voting and publicly releases the source code for community verification.
- LISABench opens Q1/2026 assessment for Web3 smart contract vulnerability discovery.
- Seven leading AI models participated, including GPT-5.2, Gemini-3-pro-preview, and Claude 4.5 Sonnet.
- Open a poll to predict the winning team and open-source the codebase on GitHub.
What is the LISABench Q1/2026 benchmark?
LISABench launched its Q1/2026 evaluation to select the most effective AI model for detecting Web3 smart contract vulnerabilities through real-world testing.
The announcement, made on January 5th, coincides with the year 2026 when the AI security sector will see its first major benchmark. The focus is on the ability to detect vulnerabilities in smart contracts, a crucial area for the cryptocurrency ecosystem due to the risks of exploitation and asset loss often stemming from code errors.
Simultaneously, LISABench launched a reward-based prediction program through community voting, aiming to attract users to follow the race and generate more feedback data on the reliability and expectations of the community for each model.
Participation models and how the community monitors
The review included 7 models: KIMI K2, DeepSeek V3.2, QWen 3 30b-a3b, GLM 4.6, GPT-5.2, Gemini-3-pro-preview, and Claude 4.5 Sonnet.
The list includes models from Moonshot AI, Alibaba Cloud, Zhipu AI, OpenAI, Google, and Anthropic, all competing on a single platform for direct comparison. This diversity of developers allows for performance comparisons based on consistent criteria, rather than relying solely on individual project claims.
Voting to predict the Q1 winner is now open. Additionally, LISABench states that the benchmark code has been open-sourced on GitHub, allowing developers to XEM and replicate the test to compare results.





