Two days ago, Google folded the limit of use of its best, Gemini 2.5 Prowhich has now been updated to move to Preview-06-05. An “updated previous” that takes another step forward to arrive soon in its final version.
Google’s goal is that Gemini 2.5 pro is Available in a couple of weeks And thus wait for Openai’s response with Chatgpt that GPT-5 continues to develop, which will be a point and apart in its business model.
From today the updated version can already be used from the calling platforms of the calls ai studio and vertex ai, just like In Gemini’s app. The only thing that demands is a subscription for use.
Google from Your blog He mentions that the previous updated Gemini 2.5 Pro continues to stand out in coding, and as proof of this is his leadership in the hardest coding tests.
It also shares the Top performance at reference points highly challenging that evaluate the capacity of reasoning, knowledge, science and mathematics of the model.
Comparative table of Gemini 2.5 Pro against other AI models
In Webdev Arena he has also been able to Overpass Claude Opus 4 to be the most powerful current model. Although in coding he has a bit to get to the experience granted by Anthropic with Claude Code.
In your ad, according to TechcrunchGoogle has commented that he has been able to Improve the structure and style of the model Previous according to the feedback collected in these weeks.
The online searches giant has given some data on Gemini 2.5 pro updated in the new previous one. A 24 -point jump in Elo in LMarena35 points on Webdevarena to lead it with 1,443 points and a higher performance in QPQA and Last examination of Humanity.
Google claims that Gemini 2.5 pro is the Reasoning Model “State-of-The-Art” (Better measurable results in a certain task or test), capable of analyzing huge data sets, code bases and documents using extensive context.
Gemini 2.5 Pro gives a lot of advantage to companies and developers at their cost, since Openai O3 exceeds at the price that the creators of Chatgpt are demanding for O4-mini. Of course, against Deepseek R1 there is nothing to do when half cost, although Gemini’s new previous one is now more powerful.
A few weeks that are very interesting for an updated previous one that has surpassed O3 and O4-Mini of OpenAi, Claude Opus 4 and Deepseek R1, and that when it arrives in its final version it is possibly waiting for it GPT-5.
