With a focus on enhanced thinking and coding skills, Google has released its Gemini 2.5 Pro language model for free users. The model, which aims to compete with top AI models like OpenAI’s o3 Mini, has a huge context window of one million tokens.
Gemini 2.5 Pro, Google’s newest voice model, is now available to users for free. While support for the iOS and Android mobile apps is anticipated shortly, the new model, which is still in the “experimental” stage, can be chosen via the drop-down menu on the Gemini website. Earlier this week, Google introduced the Gemini 2.5 Pro model for advanced Gemini users. However, because ChatGPT is taking center stage with its native picture generating capability, which is responsible for the viral Ghibli image fad, the tech giant has chosen to make the update available to free users as well.
Gemini 2.5 Pro is a reasoning model, similar to DeepSeek R1 or OpenAI’s o3 Mini. In contrast to pre-trained models or GPTs, reasoning models strive to replicate human-like reasoning skills in order to improve accuracy and performance.
What’s new with Gemini 2.5 Pro?
Google claims that Gemini 2.5 Pro is strong at reasoning and coding capabilities-related tasks. The latest language model also leads in common coding, math and science-related tasks in benchmarks such as Humanity’s Last Exam and LMArena.
Google has historically lagged behind Claude and OpenAI in terms of coding skills, but with Gemini 2.5 Pro, the corporation has placed a significant wager in this area by including a huge context window of one million tokens. With a 1 million context window, or about 7,50,000 words, the context window is the maximum quantity of text that the chatbot can process at once.
In contrast, OpenAI’s o3 Mini has a 2,00,000 context window, whilst Claude’s most recent 3.7 Sonnet model has a 5,00,000 token context window.
Google demonstrated how Gemini 2.5 Pro could create a game using just one line of text in its demo film. Once more, Google uses a variety of benchmarks to illustrate the Gemini 2.5 Pro’s programming capabilities, but these benchmarks are subject to change when new models are introduced, and actual performance is probably going to vary depending on the use case of the end user.
In order to manage increasingly complicated issues and facilitate context-aware agents, Google further stated that it is integrating these features directly into all of its models.
AI race at a boiling point:
When OpenAI released its GPT-3.5 model in late 2022, it started the AI competition. Since then, practically all of the big tech companies—including Microsoft, Google, and Meta—have placed a laser-like concentration on creating their own AI products.
It appears that the tech giants are shifting their focus from pre-trained models to reasoning models, which are also used to fuel AI agents like deep search.
With its DeepSeek R1 and DeepSeek V3 models, China’s DeepSeek disproved the notion that US-based AI companies were ahead of the curve until recently. These models were not only constructed at a fraction of the cost (despite import restrictions on AI chips), but they also competed with the most advanced Western AI models in terms of competency.
Earlier this year, when DeepSeek’s popularity peaked, panic engulfed the US stock market, wiping off millions. In an effort to demonstrate their dominance in the AI race, US tech giants have been launching ever-newer models ever since. DeepSeek is also apparently preparing to release a new model shortly.
