google gemini model-release ai-tools developer-tools

Google Launches Gemini 2.5 Pro Deep Think — New Benchmark Leader

Google's Gemini 2.5 Pro with Deep Think mode tops science and reasoning benchmarks, using parallel thinking to outperform GPT-5.5 and Fable 5.

AI Tutorials · 25 June 2026 · 2 min read

Quick answer

Google launched Gemini 2.5 Pro with Deep Think on June 22, 2026. The model scores 82.4% on GPQA Diamond and 89.8% on MMLU-Pro — the highest of any publicly available model — by exploring multiple reasoning paths in parallel before answering. It is available now in the Gemini app.

product-launch

Google Launches Gemini 2.5 Pro Deep Think — New Benchmark Leader

Google has launched Gemini 2.5 Pro with Deep Think, a new reasoning mode that makes it the top-performing AI model on science and graduate-level reasoning benchmarks. The release, which rolled out on June 22, represents the most capable publicly available model Google has ever shipped.

What Deep Think Does Differently

Most AI models generate a single chain of reasoning and commit to it. Deep Think takes a different approach: it explores multiple hypotheses in parallel, revises and combines them, then settles on the strongest answer. Google calls this “parallel thinking,” and it is visible to users — you can see thought summaries showing which paths the model considered and why it chose its final response.

The result is a model that excels at problems requiring careful deliberation. On GPQA Diamond, a benchmark of graduate-level physics, chemistry, and biology questions, Gemini 2.5 Pro Deep Think scored 82.4% — ahead of Anthropic’s Fable 5 at 79.1% and OpenAI’s GPT-5.5 at 76.3%. On MMLU-Pro, a broad knowledge benchmark, it scored 89.8%, the highest of any publicly available model.

Where It Still Trails

The model is not the best at everything. Anthropic’s Fable 5 still leads on software engineering benchmarks, scoring 88.6% on SWE-bench Verified compared to Gemini’s lower marks in that category. GPT-5.5 maintains an edge in creative writing and conversational naturalness. Each frontier model now has a clear speciality rather than dominating across the board.

What This Means for You

If you use AI for research, study, or technical problem-solving, Deep Think is worth trying. It is particularly strong at breaking down complex scientific questions, working through multi-step maths problems, and analysing code.

To use it, open the Gemini app, select Gemini 2.5 Pro from the model picker, and toggle “Deep Think” on in the prompt bar. It is available to Gemini Advanced subscribers.

This launch arrives during the most competitive stretch in AI history. With GPT-5.6 expected any day and Anthropic rapidly expanding its research team, the gap between leading models continues to narrow. For everyday users, the real winner is the growing range of capable tools to choose from — whether you need deep reasoning from Gemini, coding help from Claude, or creative writing from ChatGPT.

Frequently asked questions

What is Gemini 2.5 Pro Deep Think?

Deep Think is an extended reasoning mode for Google's Gemini 2.5 Pro model. Instead of generating one answer immediately, the model explores multiple hypotheses in parallel, revises and combines them, then selects the best response. It is designed for complex problems in science, maths, and coding.

How do I use Gemini Deep Think mode?

Open the Gemini app (gemini.google.com), select Gemini 2.5 Pro from the model picker, and toggle 'Deep Think' in the prompt bar. It is available to Gemini Advanced subscribers. Use it for hard questions where careful reasoning matters.

How does Gemini 2.5 Pro Deep Think compare to GPT-5.5 and Claude?

Gemini 2.5 Pro Deep Think leads on science and reasoning benchmarks: 82.4% on GPQA Diamond versus 79.1% for Fable 5 and 76.3% for GPT-5.5. However, Anthropic's Fable 5 still leads on software engineering tasks with 88.6% on SWE-bench Verified.

Is Gemini 2.5 Pro Deep Think free to use?

Deep Think mode is available to Gemini Advanced subscribers, which costs $19.99 per month as part of Google One AI Premium. The standard Gemini 2.5 Pro model without Deep Think is available on the free tier with usage limits.

What is the context window for Gemini 2.5 Pro?

Gemini 2.5 Pro supports up to 1 million input tokens, enough to process entire codebases, full-length books, or hours of conversation history in a single session.