MCIBench is a multilingual code intelligence benchmark for evaluating large language models on problem-aligned code generation and directed code translation. It emphasizes same-problem multilingual evaluation: models are evaluated on the same programming problems across multiple mainstream programming languages, and accepted programs can be analyzed for functional correctness, runtime, and memory usage. Benchmark Overview Item Count Problems 1,489 Languages 8 Generation qid-language slots 11,912 Directed translation language pairs 56 Translation task problems 100 Original reference-complete problems across all 8 languages 332 Augmented complete problems using accepted experimental solutions 1,466 The 8 languages are C, C++, C#, Java, JavaScript, Python3, Golang, and Rust.