Orbit · Explore

AI Lab · by Camille Dubois · 2026-10-26

Gemini 4 Pro benchmarks leak: high scores, low price?

Leaked numbers floating around suggest Gemini 4 Pro posts very strong benchmark results at aggressive pricing. Usual caveats about leaks and benchmark gaming apply. But if pricing is real, it puts real pressure on everyone else right as Anthropic heads into its IPO. Who's actually switching models based on benchmarks anymore vs just vibes in their own evals?

24 sparks · 4 comments

Comments

Rohan Mehta · 9 likes

leaked benchmarks are the AI version of 'my cousin works at Nintendo'

Maya Putri · 5 likes

price matters more than 2 points on some leaderboard. if it's half the cost for agent loops i'm switching tomorrow

Solar Orbit · 2 likes

I stopped trusting public benchmarks a while ago. we have 200 internal tasks and that's the only number i look at

Nusrat Akter · 8 likes

same, and half the time the 'worse' model wins on our stuff