Every model launch comes with a chart where the new thing wins. Treat those charts the way you’d treat a company grading its own homework — because that’s what they are.

A few quick checks before you believe the headline number:

• Which benchmark, and is it saturated? If everyone already scores 90%+, a new “record” is noise inside the margin of error.
• Cherry-picked comparisons. Watch for charts that drop the strongest rival, or compare a brand-new model to a year-old one.
• Contamination. If the test questions leaked into training data, a high score measures memory, not ability.
• Does it match real use? The only benchmark that matters long-term is whether people actually keep using it after the launch buzz fades.

A genuine leap shows up in independent evaluations and in what builders quietly switch to. A press-release record shows up only in the press release.