Google DeepMind Releases Gemini 2.5 Pro With Enhanced Reasoning

What Happened

Google DeepMind released Gemini 2.5 Pro, a new large language model the company describes as its most capable to date, with particular emphasis on extended reasoning performance across coding, mathematics, and scientific tasks. The model became available to developers through Google AI Studio and the Gemini API, with broader rollout to Gemini Advanced subscribers to follow.

Key Capabilities

Gemini 2.5 Pro is built around what Google DeepMind calls a "thinking" architecture, in which the model allocates additional compute at inference time to work through complex, multi-step problems before producing a response. The approach is comparable to reasoning-focused models released recently by competitors, including OpenAI's o3 and o4-mini series.

According to Google DeepMind's published benchmark results, Gemini 2.5 Pro achieved top scores on several standard evaluations, including AIME 2025, a mathematics competition benchmark, and coding assessments such as SWE-bench Verified, which measures a model's ability to resolve real-world software engineering tasks. The company reported a score of 63.8 percent on SWE-bench Verified, placing it at or near the top of publicly reported results at the time of release.

The model also supports a one-million-token context window, allowing it to process and reason over very large volumes of text, code, or structured data within a single session.

Availability and Pricing

Gemini 2.5 Pro is accessible through Google AI Studio at no cost during an initial period, with API pricing applying to production use cases. Google has positioned the model as suitable for enterprise developers building applications that require complex reasoning, long-document analysis, or advanced code generation.

Access through Gemini Advanced, Google's consumer-facing subscription tier, is available to users subscribed to the Google One AI Premium plan.

Background

Google DeepMind was formed in 2023 through the consolidation of Google Brain and DeepMind, two of Alphabet's primary artificial intelligence research organisations. The combined entity has since released successive versions of the Gemini model family, beginning with Gemini 1.0 in December 2023 and followed by Gemini 1.5, which introduced the extended context window that the 2.5 series retains and builds upon.

The release comes during a period of rapid model iteration across the industry. OpenAI released GPT-4.1 with a one-million-token context window and the o3 and o4-mini reasoning models within recent weeks. Meta released its Llama 4 Scout and Maverick open-weight models during the same period. The pace of releases reflects intensifying competition among major AI developers for developer adoption and enterprise contracts.

Google parent company Alphabet reported cloud revenue of 12 billion dollars in its most recent quarterly earnings, with executives citing AI-related demand as a primary growth driver. Google Cloud is one of the primary commercial channels through which Gemini models are offered to enterprise customers.

What It Means in Practice

Developers accessing Gemini 2.5 Pro through the API can apply it to tasks including automated code review, long-document summarisation, complex data analysis, and multi-step scientific reasoning. The one-million-token context window means the model can ingest documents or codebases substantially larger than those supported by earlier model generations without splitting input across multiple sessions.

The reasoning architecture adds latency compared to standard generation models, as the model performs additional internal processing steps before returning output. Google DeepMind has not published specific latency figures for average use cases.

Google DeepMind has indicated that additional Gemini 2.5 variants, including versions optimised for speed and cost efficiency, are under development and expected to be released in the coming months.