Live Code Bench
LiveCodeBench is a programming ability evaluation benchmark for large language models. It continuously collects problems from competitive programming platforms such as LeetCode and Codeforces and dynamically updates its problem set to reduce the impact of training data contamination on evaluation results. The benchmark covers multiple dimensions including code generation, debugging, and execution reasoning, aiming to more objectively reflect model performance on real-world programming tasks.
Timeline (last 90 days)
Google's official model card gave Gemma 4 26b a score of 77.1% on Live Code Bench.
Live Code Bench problems are similar to LeetCode competitions, requiring implementation given an independent function signature, and almost never involve cross-file dependencies or incremental modifications.
On Qwen's model card, Qwen3.8 27b scored 61.7 on SWE-Bench Pro, 73 on Terminal Bench 2.1, and 90.3 on Live Code Bench.
Qwen3.8 Flash Next scored 84.5 on the Android World benchmark and 91.9/95.7 on Live Code Bench (both are official figures)
On Live Code Bench, Qwen3.8-27B scored 90.3 and Claude Opus 4.6 scored 88.8.
All Facts (5)
Google's official model card gave Gemma 4 26b a score of 77.1% on Live Code Bench.
50%UnverifiedLive Code Bench problems are similar to LeetCode competitions, requiring implementation given an independent function signature, and almost never involve cross-file dependencies or incremental modifications.
50%UnverifiedOn Qwen's model card, Qwen3.8 27b scored 61.7 on SWE-Bench Pro, 73 on Terminal Bench 2.1, and 90.3 on Live Code Bench.
50%UnverifiedQwen3.8 Flash Next scored 84.5 on the Android World benchmark and 91.9/95.7 on Live Code Bench (both are official figures)
50%UnverifiedOn Live Code Bench, Qwen3.8-27B scored 90.3 and Claude Opus 4.6 scored 88.8.
50%