Gemini 2.5 Pro Model with Enhanced Reasoning Now in Public Preview
Google Launches Gemini 2.5 Pro in Public Preview
Google LLC has officially launched the public preview of its advanced large language model, Gemini 2.5 Pro, which was first introduced last month. This move enables broader access for developers and businesses looking to leverage advanced AI capabilities.
Enhanced Access for Developers
Previously, developers could only interact with Gemini 2.5 Pro through a free API that had strict usage limits. Users were restricted to just 25 requests per day, at a maximum pace of five requests per minute. The new public preview version changes this dynamic significantly by providing paid tiers which allow for much higher usage limits.
Under the new structure, developers can submit up to 2,000 requests per minute without a daily cap. Moreover, Gemini 2.5 Pro has the capability to process substantial amounts of data—up to eight million tokens every minute—making it a robust option for applications serving large audiences.
Cost Structure and Pricing
Google has established a tiered pricing model for the usage of Gemini 2.5 Pro. Here are the key details:
For prompts up to 200,000 tokens:
- Input tokens: $1.25 per million
- Output tokens: $15 per million
- For prompts exceeding 200,000 tokens:
- Input tokens: $2.50 per million
- Output tokens: $15 per million
This pricing makes Gemini 2.5 Pro a more affordable alternative compared to competitors like DeepSeek-R1, while it remains pricier than Anthropic PBC’s Claude 3.7 Sonnet.
Performance and Benchmarks
Upon its initial launch, Gemini 2.5 Pro received high marks, topping the LMArena LLM benchmark by a significant margin. This benchmark is designed to assess the performance of AI models based on user feedback. Additionally, Gemini 2.5 Pro scored an impressive 86.7% on the AIME 2025, a qualifying exam for the U.S. Math Olympiad.
One of the remarkable aspects of Gemini 2.5 Pro is its ability to outperform other models optimized for reasoning without utilizing costly test-time techniques. These techniques often require significant computing power, impacting the overall cost of model deployment.
Technical Enhancements
Gemini 2.5 Pro is based on improvements made to the earlier Gemini 2.0 Pro model introduced in December. Google’s engineers have focused on enhancing both the core model and its post-training workflow—an important method for refining the output quality by utilizing additional data after initial training.
Competitive Landscape
Google’s Gemini series is intensifying competition in the AI space, primarily against established players like OpenAI. In response to this competitive push, OpenAI is set to unveil two new reasoning models within the next two weeks. According to OpenAI CEO Sam Altman, the company will introduce:
- The previously-previewed o3 reasoning model.
- A new model called o4-mini, which was not mentioned before.
Furthermore, OpenAI plans to introduce GPT-5 in a few months, which is expected to integrate several features along with the reasoning capabilities of the o3 model. This new system will be used for both free and paid versions of ChatGPT.
Google’s updated offerings and pricing structure reflect its commitment to providing advanced AI tools while fostering an environment for developers to experiment and innovate. The ongoing enhancements to both Gemini and OpenAI models will likely continue to shape the future of artificial intelligence.
