Gemini 2.5 Pro Thinking - First Look

Показать описание

In this video, we take a first look at Google's newly released Gemini 2.5 Pro — their most advanced AI model yet. With impressive benchmark results, state-of-the-art reasoning, and next-level coding capabilities, Gemini 2.5 Pro might just be a game-changer. Watch as we explore its chain-of-thought reasoning, test its coding skills, and see how it stacks up against other frontier models.

LINKS:

RAG Beyond Basics Course:

Let's Connect:

Signup for Newsletter, localgpt:

Рекомендации по теме

Комментарии

It's out in Ai Studio now and works better than the Gemini web interface. I think this is such a big step up and the 1 million context window is nice.

FactsNoCare

To answer you question at around 2:50 - What they mean is the Grok and O3 often cheat on their benchmarks by getting their model to answer 10 times, and picking the answer using majority vote - like no user actually does this, but they do it just to show their model as SOTA.

TheVistastube

These AIS are getting crazy good and its really exciting especially for no code geeks

newstodayyt

Thanks for the great and quick summary. Great work!

Ikbeneengeit

Cant believe it’s free, 1 mill context and SOTA lol, I love google

MugiwaraNoDeji

0:42 The offline Internet dinosaur game! That's it. I'm adding Gemini 2.5 to my P2P multi-agent list

LobbyXReddit

Thanks! I just subscribed. I'm excited to try it out with Vertex AI.

VictorSharamet

This is good. Could you also test the models with their tool or function calling capabilities as part of your testing tasks? Similar to how Anthropic validates their models for τ-benchmark (tau).

KishanLal-sk

02:40 It literally says in the same sentence: "Test time techniques... like majority voting" 🤣

MrMiguelChaves

interesting:) and has structured output:)

micbab-vgmu

i have heard test time compute first when it comes to the arc challange. There it stands for the model using the given example to train on in other words training on the example you gave it to hopfully adjust itself to your question instead of being a jack of all traits it optemizes itself on your problem while you asking it a question.

VR_Wizard