For the latest updates and exclusive contents of the industry’s best AI application, join the daily and weekly newsletter. Learn more
Only a few months after the release of Gemini 2.0 Deepseek,,, Google The “most intelligent model” has yet announced Gemini 2.5, which is still inferred and improved performance and accuracy.
Gemini 2.5 has been three months since Google introduced reasoning and agent use three months after the previous release of Gemini 2.0, the most intelligent model family. This new model is provided as a Gemini Advanced user of the Gemini 2.5 PRO in Google’s AI Studio. It can be used soon in Vertex AI.
KAVUKCUOGLU, CTO of Google Deepmind, said in IN Blog Gemini 2.5 indicates the next step of the goal of Google creating the ability to “smarter and more inferences”.
Kavukcuoglu said, “Now, through Gemini 2.5, we have achieved a greatly improved basic model and a new level of improved performance after training.“ We build these thinking ability directly to all models, so we can handle more complex problems and support more competent situation recognition agents. ”
More context and understanding
Like the Gemini 2.0 and Gemini 2.0 flash accidents, Gemini 2.5 Pro is “thought” before responding. The new model can handle multimodal inputs in text, audio, images, videos and large data sets. Gemini 2.5 Pro can understand the entire code repository for coding projects.
Gemini 2.5 Pro provides some of the largest context windows available for Gemini’s experimental models. It will be shipped with a million token context window, but will soon expand to 2 million tokens. Google AI Studio Product Manager Logan Kilpatrick posted Gemini 2.5 Pro on X as “the first experimental model with a high speed limit + claim.”
Google plans to launch a price set for the Gemini 2.5 model soon.
Improving coding and reasoning performance
Google said that this model is leading the high -quality benchmark test. The company said, “Gemini 2.5 Pro said,“ We are leading the game and science benchmarks such as GPQA and AIME 2025. KAVUKCUOGLU has a state -of -the -art 18.8%cutting -edge score in the model without the use of tools in the last human test.
The Gemini 2.5 Pro performs the coding task strongly and takes a better score than the Gemini 2.0 on a specific benchmark. Google said in a new model, “It is excellent for creating a visually attractive web app and agent code application and code conversion and editing.”
A more competitive market
The Gemini 2.5 Pro enters the reasoning model fight in an environment that is much larger than the December Gemini 2.0. The launch of DeepSeek’s inference large language model (LLM) showed that strong models can be performed well in some of the cost of training and calculation. DeepSeek also shows that open source models can compete with more closed source LLM. OpenaiO1 and O3 models.
In addition to the constantly expanded model offering DeepSeek, Google must compete with Openai’s reasoning model. Openai’s latest models are GPT-4.5, not reasoning models, but the company is still expected to develop more reasoning models.
Gemini 2.5 The second new model of Google this month. In March, the company announced the latest version of Gemma 3, a small language model.It provided 128,000 token context models and was best for use on devices on the go.
