Google has transitioned its Gemini 2.5 Pro large language model into public preview, allowing broader access for developers via Google AI Studio and the Gemini API.
Highlights
The move follows its initial launch in March, which offered experimental access under limited usage caps. In response to growing demand, Google has raised rate limits and introduced a two-tier pricing structure designed to offer competitive rates within the AI model market.
With expanded access, developers can now integrate Gemini 2.5 Pro into their applications more freely using Google AI Studio.
While the model is not yet available on Vertex AI, Google has confirmed that integration is planned for the near future. The public preview reflects a shift from early experimental stages toward wider adoption, driven by strong interest from the development community.
The pricing model is structured in two tiers based on token usage. For usage under 200,000 tokens, the cost is $1.25 per million input tokens and $10 per million output tokens.
This includes input across multiple formats, such as text, images, audio, and video. For usage beyond that threshold, pricing increases to $2.50 per million input tokens and $15 per million output tokens.
Google continues to offer free access to an experimental version of Gemini with reduced rate limits.
Google states that its pricing remains competitive compared to other models in the industry. For instance, Anthropic’s Claude 3.5 Sonnet is priced at $3 per million input tokens and $15 per million output tokens.
OpenAI’s o1 model costs $15 per million input tokens and $60 per million output tokens, although discounts are available for cached inputs, reducing that figure to $7.50. By contrast, Gemini 2.5 Pro maintains lower pricing across its tiers.
| Model | Input Token Cost (per 1M tokens) |
Output Token Cost (per 1M tokens) |
Free Tier | Notes |
|---|---|---|---|---|
| Gemini 2.5 Pro | $1.25 (first 200k) $2.50 (after) |
$10.00 (first 200k) $15.00 (after) |
Yes (limited rate) | Includes multi-modal inputs (text, image, audio, video) |
| Claude 3.5 Sonnet | $3.00 | $15.00 | No | Flat pricing across usage |
| OpenAI o1 (GPT-4-turbo) | $15.00 ($7.50 cached) |
$60.00 | No | Discounted cached input tokens |
While Gemini 2.5 Pro is described as Google’s most advanced model to date, the company has not disclosed specific architectural changes or performance enhancements over earlier versions.
Nonetheless, the broader rollout and relatively accessible pricing indicate Google’s intention to encourage widespread use among developers and organizations incorporating generative AI into their operations.
Enhanced Reasoning and Multimodal Capabilities
Gemini 2.5 Pro includes improved reasoning abilities, enabling it to break down complex tasks into logical steps before responding.
This methodical approach contributes to higher accuracy, particularly in domains such as mathematics, science, and programming.
Its built-in multimodal support allows it to interpret and analyze text, audio, image, and video inputs, offering robust functionality for use cases that span diverse media formats.
Support for Large Contexts
One of the model’s key features is its extended context window, currently supporting up to 1 million tokens, with future plans to double this to 2 million.
This enhancement enables the model to process substantial amounts of information, such as entire codebases or long-form documents, supporting more detailed and context-aware responses.
Benchmark Performance
Gemini 2.5 Pro has shown strong results in industry evaluations. It leads the LMArena leaderboard by a wide margin and scored highly on standardized benchmarks, achieving 86.7% on the AIME 2025 and 84.0% on the GPQA Diamond benchmark.
These outcomes reflect its capabilities in handling complex reasoning and problem-solving tasks.
Developer Access and Integration
Currently accessible via Google AI Studio and the Gemini app for Advanced users, Gemini 2.5 Pro will soon be integrated into Vertex AI.
The staged rollout is designed to provide flexible access for both individual developers and enterprise users aiming to incorporate generative AI into their products and workflows.
