Google DeepMind Releases Gemini 2.5 Pro Preview to Developers

What Happened

Google DeepMind released Gemini 2.5 Pro, its most capable multimodal AI model to date, to developers through the Gemini API on April 9, 2025. The release makes the model broadly accessible for the first time after an earlier limited preview, and positions it as Google's flagship offering for complex reasoning, coding, and long-context tasks.

Background

Gemini 2.5 Pro is the latest in Google DeepMind's Gemini family of large language models, which the company has developed as its primary AI platform for both consumer and enterprise applications. The model was first previewed in March 2025, when Google made it available to a restricted set of users and benchmark evaluators. At the time of that preview, Google reported that Gemini 2.5 Pro achieved top scores on several widely followed AI benchmarks, including coding evaluations such as SWE-bench Verified and reasoning tests such as GPQA Diamond.

Google DeepMind operates as the consolidated AI research division of Alphabet, formed in 2023 through the merger of Google Brain and DeepMind. The division is responsible for both foundational model research and the deployment of AI products across Google's consumer and cloud services.

Model Capabilities

Gemini 2.5 Pro supports a context window of one million tokens, allowing it to process and reason over large volumes of text, code, images, audio, and video within a single prompt. Google has described the model as a "thinking" model, meaning it applies an internal chain-of-thought reasoning process before producing a final response.

The model is accessible through Google AI Studio, which provides a free-tier interface for developers, and through the Gemini API, which supports paid access with higher rate limits. Google has also announced plans to make Gemini 2.5 Pro available through Vertex AI, its enterprise cloud platform on Google Cloud.

On the Chatbot Arena leaderboard maintained by researchers at the University of California, Berkeley, Gemini 2.5 Pro ranked first at the time of its broader release, based on human preference ratings across a large volume of head-to-head model comparisons.

Pricing and Access

Google set the API pricing for Gemini 2.5 Pro at $1.25 per million input tokens for prompts under 200,000 tokens, and $2.50 per million input tokens for prompts exceeding that threshold. Output tokens are priced at $10.00 per million. These figures place the model in a competitive range with comparable frontier models from OpenAI and Anthropic.

Developers using Google AI Studio can access Gemini 2.5 Pro without charge up to defined rate limits, a structure Google has used with previous model releases to encourage adoption among individual developers and researchers before they scale to paid tiers.

Competitive Context

The release comes during a period of rapid model deployment across the AI industry. OpenAI released its o3 and o4-mini reasoning models in April 2025. Anthropic's Claude 3.7 Sonnet, released earlier in the year, introduced a hybrid reasoning mode that the company positioned as a competitor to Google's thinking model approach. Meta released its Llama 4 open model series in April 2025, expanding the range of options available to developers who prefer open-weight models.

Google has not announced a specific timeline for a full production release of Gemini 2.5 Pro beyond the current preview designation, though the company has indicated that Vertex AI availability will follow the initial API rollout.

What Happens Next

Google has stated that Gemini 2.5 Pro will become available on Vertex AI for enterprise customers in the coming weeks, extending access to organizations that require data residency controls and enterprise support agreements.