TLDRs;
- Google launches Deep Think AI exclusively for Gemini Ultra subscribers at $250 monthly.
- Deep Think outperforms OpenAI’s o3 and xAI’s Grok 4 in benchmark tests.
- The model leverages multimodal and parallel reasoning for superior problem-solving.
- Public access version hits bronze-level Olympiad standards in internal testing.
Google has officially released Deep Think, the latest and most advanced AI model for its Gemini platform. Available exclusively to Gemini Ultra subscribers, the cutting-edge reasoning model is priced at $250 per month.
The move marks a major step forward in Google’s AI roadmap, following the model’s initial unveiling at the company’s I/O developer conference earlier this year.
Described as a variation of the model that achieved gold-medal performance at the International Mathematical Olympiad, the public version of Deep Think is optimized for faster and more practical everyday use. Internal tests show it would perform at a bronze-medal standard in the same Olympiad, offering a strong balance between capability and speed.
Currently, Deep Think is accessible only to subscribers on the highest AI plan. While users can access it within the Gemini app, prompt usage will be capped per day, though Google has yet to confirm the specific limits.
Starting today, we're offering Deep Think in the Gemini app for Google AI Ultra subscribers, and we're giving a collection of mathematicians access to the full version of the Gemini 2.5 Deep Think model that achieved gold medal🏅 level performance in the recent IMO competition.… pic.twitter.com/pfvefrOQS5
— Jeff Dean (@JeffDean) August 1, 2025
Parallel Thinking and Multimodal Input
Deep Think’s standout feature is its “parallel thinking” architecture. This allows the model to generate and consider multiple ideas simultaneously, a leap from traditional sequential models. According to Google, this not only accelerates problem-solving but significantly boosts reasoning depth.
The AI is also multimodal, capable of processing inputs in text, image, and audio formats. This positions it well for complex real-world applications across disciplines, from academic research to enterprise problem-solving. Early access testers reportedly praised its intuitive handling of multi-format tasks.
These capabilities build upon recent additions to Google’s broader AI ecosystem. Last month, the company rolled out Gemini 2.5 Pro and Deep Search within its AI Mode chatbot, signaling a strategic shift in how Google integrates AI into search and daily user engagement.
Deep Think Outperforms Rivals
According to performance tests conducted by Google, Deep Think scored higher than top-tier models from competing AI developers. It outperformed OpenAI’s o3, xAI’s Grok 4, and even Google’s own Gemini 2.5 Pro across standard benchmarks like Humanity’s Last Exam and LiveCodeBench.
This performance boost underscores Google’s intent to lead the competitive frontier in advanced AI development. The benchmarking results will likely intensify the ongoing rivalry between Google, OpenAI, and Elon Musk’s xAI, as each firm races to dominate the next phase of AI evolution.
Despite its high cost, the rollout is seen as a necessary proving ground for Google’s most capable technology before wider deployment. A more advanced gold-medal version is currently restricted to a small group of mathematicians and academics for further testing and research.
The Future of Google AI
Deep Think is part of a broader vision where AI becomes central to how users interact with Google products. With AI Mode expanding, Deep Search being integrated into search functions, and features like AI-powered phone calls through Project Astra in development, Google is betting on a future where AI tools are always-on and omnipresent.
As Deep Think matures and user feedback grows, it’s likely that more refined or affordable versions could reach broader segments. For now, Gemini Ultra subscribers will be the first to explore the limits of what Google’s most powerful model can do.
