Google DeepMind Releases Gemini 2.5 Pro With Enhanced Reasoning

What Happened

Google DeepMind released Gemini 2.5 Pro Preview, a new large language model with enhanced reasoning capabilities, making it available to developers and enterprise users through Google AI Studio and the Gemini API. The release positions the model directly against OpenAI's o3 and Anthropic's Claude 3.7 Sonnet in the competitive segment of reasoning-capable frontier models.

Key Capabilities

Gemini 2.5 Pro is designed to work through complex, multi-step problems before producing a final answer, a design pattern that AI developers refer to as chain-of-thought or extended thinking. Google DeepMind stated the model achieves strong performance on tasks involving coding, mathematics, and scientific reasoning.

At the time of release, Gemini 2.5 Pro ranked first on the Chatbot Arena leaderboard maintained by researchers at LMSYS, a widely cited independent evaluation platform where human raters compare model outputs. The model also recorded leading scores on several coding benchmarks, including SWE-bench Verified, which tests models on real-world software engineering tasks drawn from open-source repositories.

The model supports a context window of one million tokens, matching the capacity introduced in earlier Gemini 1.5 versions. Google DeepMind stated that a two-million-token context option is planned for a subsequent release.

Pricing and Access

Gemini 2.5 Pro Preview is available through Google AI Studio at no cost during the preview period, subject to rate limits. Paid API access is priced at $1.25 per million input tokens for prompts under 200,000 tokens, and $2.50 per million input tokens for longer prompts. Output tokens are priced at $10 per million. These rates are consistent with pricing tiers Google has applied to earlier Gemini 1.5 Pro releases.

Enterprise access is available through Google Cloud's Vertex AI platform. Google did not specify a date for when the preview label will be removed and the model will move to general availability.

Background

Google DeepMind was formed in 2023 through the consolidation of Google Brain and the original DeepMind research organisation. The combined entity reports to Demis Hassabis, who serves as chief executive. The Gemini model family, first released in December 2023, serves as the primary large language model line for Google's consumer products, developer tools, and cloud services.

The Gemini 2.5 Pro release follows Gemini 2.0 Flash, a smaller and faster model released earlier in 2025, and comes as Google prepares for its annual Google I/O developer conference, scheduled for May 2025. Google has historically used I/O to announce major product and platform updates.

The reasoning model segment has grown rapidly in early 2025. OpenAI released o3 and o4-mini in April 2025, and Anthropic updated Claude 3.7 Sonnet with an extended thinking mode in February 2025. Google's release of Gemini 2.5 Pro brings the company's flagship model into direct competition in that category.

What It Means in Practice

Developers accessing the model through the API can use it for tasks that require sustained logical reasoning, such as generating and debugging code, solving mathematical proofs, and analysing lengthy documents. The one-million-token context window allows the model to process book-length texts, large codebases, or extended conversation histories within a single prompt.

Google did not announce multimodal input capabilities beyond text for this preview release, though earlier Gemini models have supported image, audio, and video inputs. The company indicated that additional capabilities will be detailed at upcoming developer events.

Google has scheduled its Google I/O developer conference for May 2025, where further announcements about the Gemini 2.5 model family and associated developer tools are expected.