Google DeepMind Releases Gemini 2.5 Pro With Enhanced Reasoning

What Happened

Google DeepMind released Gemini 2.5 Pro, its latest and most capable large language model, featuring enhanced reasoning capabilities that the company says represent a substantial step forward from its predecessor. The release positions Gemini 2.5 Pro as Google's flagship model for tasks requiring multi-step reasoning, coding, and complex problem solving.

Key Capabilities

Gemini 2.5 Pro introduces what Google DeepMind describes as a "thinking" mode, in which the model reasons through problems before producing a final response. This approach mirrors techniques used in competing reasoning-focused models, including OpenAI's o-series and Anthropic's extended thinking mode introduced with Claude 3.7 Sonnet.

The model supports a context window of up to 1 million tokens, allowing it to process large volumes of text, code, or documents within a single prompt. Google DeepMind has indicated the context window may expand to 2 million tokens for select users.

On publicly reported benchmarks, Gemini 2.5 Pro reached top scores on several coding and mathematics evaluations, including the AIME mathematics competition test and the SWE-bench coding benchmark, which measures a model's ability to resolve real-world software engineering tasks.

Background

Google DeepMind is the consolidated AI research division of Alphabet, formed in 2023 through the merger of Google Brain and DeepMind. The division is responsible for developing the Gemini model family, which launched in late 2023 as Google's response to the rapid commercial adoption of OpenAI's GPT-4 series.

Gemini 2.5 Pro follows Gemini 2.0, released earlier in 2025, which itself introduced multimodal capabilities and improved performance over the original Gemini 1.5 Pro. The 2.5 designation reflects an incremental generational update rather than a full architectural rebuild, according to materials published by the company.

Google DeepMind makes Gemini models available through Google AI Studio for developers and through the Gemini API, as well as integrating them into Google's consumer and enterprise products including Google Workspace and the Gemini assistant.

Benchmark Performance and Context

Gemini 2.5 Pro's reported benchmark results place it at or near the top of several publicly tracked leaderboards at the time of release. The model scored highly on MMLU Pro, a test of professional-level knowledge across domains, and on LiveCodeBench, which evaluates coding ability using problems released after a model's training cutoff to reduce the effect of data contamination.

Benchmark comparisons in the AI industry carry inherent caveats. Results can vary depending on evaluation methodology, prompt formatting, and whether models are tested in their standard or extended reasoning modes. Google DeepMind published its evaluation methodology alongside the release.

The release arrives within weeks of OpenAI publishing its o3 and o4-mini reasoning models, and Meta releasing the Llama 4 model family with multimodal capabilities. The cluster of major releases reflects an accelerating pace of model development across the industry's leading laboratories.

Availability and Pricing

Gemini 2.5 Pro is available in Google AI Studio and through the Gemini API. Pricing for API access is structured around token volume, with input and output tokens billed separately. Enterprise pricing is available through Google Cloud's Vertex AI platform, where the model is also accessible.

Consumer access to Gemini 2.5 Pro is available to Google One AI Premium subscribers, who receive priority access to the company's most capable models through the Gemini web and mobile interfaces.

What Happens Next

Google DeepMind has indicated that broader availability of the 2-million-token context window and additional capability updates to the Gemini 2.5 family are planned for subsequent releases in 2025.