Gemini 2.5 Deep Think (blog.google)

Article summary

Google has released Deep Think in the Gemini app for Google AI Ultra subscribers, a tool that uses parallel thinking to solve complex problems. Deep Think is a variation of the model that achieved the gold-medal standard at the International Mathematical Olympiad and can help with tasks such as coding, scientific discovery, and mathematical reasoning. The model uses parallel thinking techniques to generate many ideas at once and consider them simultaneously, before arriving at the best answer. This release represents a significant step forward in Google's mission to build more helpful and capable AI.

Main themes

• AI capabilities
• Parallel thinking
• Gemini app
• Google AI Ultra
• Mathematical reasoning
• Coding and development

What commenters say

• The approach of using multiple agents in parallel to achieve better results is not new, but it is now making a difference with powerful reasoning models.
• The use of parallel thinking techniques is a simple yet effective approach that has taken a long time to be implemented.
• Some commenters are skeptical about the benefits of running multiple instances of the same model versus running different models with different architectures.
• The cost of using Deep Think, at $250 per month, is prohibitively expensive for some users.
• Local LLM enthusiasts may be able to replicate the results of Deep Think using their own models and hardware, potentially at a lower cost.
• The comparison between Deep Think and other models, such as Grok 4 Heavy, is not entirely fair due to differences in benchmarking and tool usage.
• Some users are frustrated with the aggressive marketing of Gemini and the frequent prompts to use the service.
• The integration of AI into Google Workspace is seen as useful and relevant by some users, but annoying and cluttered by others.