Lingo-Judge

Lingo-Judge is the grading standard used to score answers given by autonomous driving AI. Just as a teacher grades an exam paper, autonomous driving AI is asked how it should judge a given situation, and its answer is then scored against this standard.

This scoring is used within LingoQA, a reasoning test for autonomous driving. Different companies' autonomous driving models take the same test, get scored under the Lingo-Judge criterion, and are then compared against one another. For example, an article might report that NVIDIA's autonomous driving model scored several points higher than other well-known models under this criterion.

However, this score is closer to a relative comparison figure than an absolute report card. Because the score can shift if the grading criteria or the composition of the test questions change, it should be read alongside the fact that it comes from the announcing company's own testing.

Articles present it in a form such as "17.0 points higher than Qwen2.5-VL 72B and 23.2 points higher than GPT-4o under the Lingo-Judge metric." Here, the score does not represent an absolute perfect-score rating but rather relative superiority within this particular scoring method, and since it is a result measured directly by the announcing company, it should be distinguished from third-party verification.