METR FINDS DEEPSEEK AND QWEN MODELS LAG FRONTIER AI BY ABOUT 6 MONTHS

Summary from the TLDR AI newsletter

The evaluation tested autonomous capabilities across software engineering tasks, finding mid-2025 Chinese models perform similarly to late 2024 frontier models.