Gemini 2.5 Deep Think is a new model for Google AI Ultra subscribers. It's a version of the model that achieved a gold-medal standard at the IMO, designed to solve complex problems with parallel "thinking time".
This new "Deep Think" model is now rolling out in the Gemini app for Google AI Ultra subscribers. It uses extended, parallel "thinking time" to let the AI explore different hypotheses before giving an answer, which is great for tackling complex problems in coding, math, or creative planning.
It's a faster version for daily use, but Google is also giving the full, powerful IMO-version to select mathematicians to advance research.
Anyway, who still thinks they're smarter than AI? :))
It's a bit sad that Deep Think is only available for Ultra subscribers at $250 a month, that's a steep price for most people. But the results are impressive: on Humanity's Last Exam, the model scored 34.8%, which is much higher than Grok 4 (25.4%) and OpenAI's o3 (20.3%). On LiveCodeBench, the gap is even bigger, 87.6% versus 79% and 72%. For those who can afford it, this is a whole new level
Wow, the idea of actually being able to find anything on the internet blows my mind—this is going to save me hours hunting for info for my uni projects. How fast is it?
Whoa, organizing all the world’s info in one place? That’s wild. I can never find what I need—this could save me hours hunting stuff down! How fast does it update?
Replies
Hi everyone!
So this is the one that recently won a gold medal at the IMO for Google.
This new "Deep Think" model is now rolling out in the Gemini app for Google AI Ultra subscribers. It uses extended, parallel "thinking time" to let the AI explore different hypotheses before giving an answer, which is great for tackling complex problems in coding, math, or creative planning.
It's a faster version for daily use, but Google is also giving the full, powerful IMO-version to select mathematicians to advance research.
Anyway, who still thinks they're smarter than AI? :))
It's a bit sad that Deep Think is only available for Ultra subscribers at $250 a month, that's a steep price for most people. But the results are impressive: on Humanity's Last Exam, the model scored 34.8%, which is much higher than Grok 4 (25.4%) and OpenAI's o3 (20.3%). On LiveCodeBench, the gap is even bigger, 87.6% versus 79% and 72%. For those who can afford it, this is a whole new level
BestPage.ai
Wow, the idea of actually being able to find anything on the internet blows my mind—this is going to save me hours hunting for info for my uni projects. How fast is it?
BestPage.ai
Whoa, organizing all the world’s info in one place? That’s wild. I can never find what I need—this could save me hours hunting stuff down! How fast does it update?