编程AI性能追踪器

Models are ranked using a benchmark-driven score from 0 to 100 where benchmark performance (90%) from LMArena, MMLU, GPQA, HumanEval, SWE-bench, and 15+ standardized evaluations is the primary signal. Capabilities and context window serve as tiebreakers (10%). Scores update hourly from live API data across 290+ coding models.

The 24h column shows how many positions a model moved up or down in the last 24 hours, while the 7d column shows the change over the past week. Green values with a + indicate rank improvements, red values indicate drops, and 0 means the rank was unchanged.