What is Deep Reasoning?
Deep reasoning (sometimes called "thinking mode" or "extended thinking") is a special mode where the AI takes extra time to work through complex problems step by step before giving you an answer. Instead of responding instantly, it reasons through the logic, considers edge cases, and checks its own work. The result is significantly more accurate answers for hard problems, at the cost of a longer wait.
Common use cases- Solving multi step math, logic, or coding problems that require careful reasoning
- Analyzing complex documents where the AI needs to cross reference multiple sections
- Making decisions that involve weighing several factors or trade offs
Step 1: Select Deep Think
Choose the Deep Think model in the Gemini model picker (Ultra only)
Step 2: Enter your problem
Provide a complex math, science, or engineering question
Step 3: Deliberation
Gemini reasons through the problem with iterative multi hypothesis exploration
Step 4: Detailed answer
Delivers a thorough, step by step solution after 30 seconds to several minutes
0.4%
Humanity's Last Exam (hardest AI test)
0.6%
ARC-AGI-2 (reasoning test)
0
Codeforces rating (top 0.1% of coders)
What do these benchmarks mean?
Deep Think (Gemini)
Approach
Iterative multi hypothesis
Minimum plan
Ultra ($249.99/mo)
Context window
1M tokens
Strength
Gold medal IMO, benchmarks
o3 (OpenAI)
Approach
Chain of thought RL
Minimum plan
Plus ($20/mo)
Context window
128K tokens
Strength
Competitive math/science
Extended Thinking (Claude)
Approach
Extended internal deliberation
Minimum plan
Pro ($20/mo)
Context window
200K tokens
Strength
Writing + reasoning