Google today announced the latest generation of its artificial ielligence models with the name Gemini 3 iroduced; According to this company, the model is considered to be the most accurate and powerful model in the history of Google and is considered a direct replaceme and competitor of GPT-5. The Pro version of this model is available in Google products from today, and the Deep Think version, focusing on reasoning, has started the test phase.
Google today announced that his new artificial ielligence family with the name Gemini 3 has made it available to users. This company explained that Gemini 3 is the “smartest model” in the history of Google and its accuracy has made a significa leap compared to the previous generation. According to Google, this model is a direct response to GPT-5 is considered; A model iroduced by OpenAI a while ago.
A gia leap in multimodal capabilities
Google expressed The Gemini 3 is the company’s flagship model for multimodality perception and “naturally” processes text, images, video and audio. This model is also the most powerful operating system (Ageic) and Vibecoding Google is considered.
According to Google, supply Gemini 3 Pro In the form of a preview, it is the beginning of “Asr Jaminai 3” and has been activated in the product collection of this company since today.
Iroducing Gemini 3 Deep Think
Google coinues announced that model Gemini 3 Deep Think has also developed A model that is capable of deeper reasoning and surpasses Gemina 3 Pro. This version will be released first to safety testers and then to Google AI Ultra subscribers.
Gemini 3 performance in benchmarks
Google explained Gemini 3 Pro has improved in all benchmarks compared to version 2.5. This model now:
in LMArena Leaderboard with 1501 Elo It is at the top.
in the test Humanity Last Exam to score 37.5% Arrived without tools.
in GPQA Diamond Score 91.9% and has appeared at the level of “doctoral reasoning”.
in MathArena Apex with 23.4% It has set a new standard for mathematics.
In the multifaceted field, Gemini 3 Peru:
Its coe accuracy is also in SimpleQA Verified to 72.1% has arrived
Google claimed that this model provides “things you need to hear” and acts like a “real thought partner” and provides users with new ways to understand and express coe.
Deep Think; Deeper digestion of information
The Deep Think version performs even better in benchmarks and:
41% In Humanity Last Exam
93.8% In GPQA Diamond
and 45.1% In ARC-AGI-2
has obtained; The results show its ability to solve new and complex problems.
Practical applications of Gemina 3
Google explained Gemina 3 has been designed from the beginning to produce all kinds of coe in any format; From text and image to video, sound and code.
The company gave the example that a user could provide the model with handwritten family cooking instructions, from which a A complete and shareable cookbook compile Users can also submit articles, long videos, or tutorials to Gemina to share them with Flashcards, images or educational formats no longer convert
Google artificial ielligence in search and software developme
Ability AI Mode Google Search is now enabled on Gemini 3 and can create visual layouts, ieractive tools or real-time simulations.
Google announced which Gemina 3 is the best model in the field of Vibecoding and age-orieed. This model:
in WebDev Arena Score 1487 Elo
in Terminal-Bench 2.0 Score 54.2%
and in SWE-Bench Verified Score 76.2%
has registered
Gemini Age feature
Google expressed that Gemini 3 Pro is the main engine capability Gemini Age is An age that can sort emails, perform various researches, or complete the process of booking a travel ticket instead of the user.
Supply and access
Google announced Gemini 3 Pro has been activated for all users on the Gemini app from today. Subscribers Google AI Pro and AI Ultra In the United States, they can also experience Gemini Age and AI Mode based on Gemini 3 Pro.
Finally, by iroducing Gemini 3, Google has taken an importa step in the competition of advanced artificial ielligence models. Higher accuracy, extensive multimodality capability, advanced reasoning, and age-orieed capabilities make this model a direct competitor to GPT-5. With the release of the Pro version and the start of the Deep Think version test, Google is trying to establish its position in the new generation of basic models; The generation that is going to play a promine role in coe creation tools, software developme and future search experiences.
