Google has introduced an updated version of its reasoning-focused AI model, Gemini 2.0 Flash Thinking, as part of its experimental lineup. Initially launched in December 2024, this release marks a significant enhancement in the model's capabilities, with the new experimental variant (Exp-01-21) becoming available on January 22, 2025. Below is a breakdown of the key aspects of this update:
The New Feature
The Gemini 2.0 Flash Thinking update introduces several advanced features aimed at improving reasoning and usability:
- 1 Million Token Context Window: This is a substantial increase from the previous 32,000 tokens, enabling the model to handle extensive and complex inputs such as entire codebases or large academic datasets.
- Native Code Execution Support: Users can now execute code directly within the model's responses, enhancing its utility for programming and computational tasks.
- Longer Output Token Generation: The model can produce more extended responses, which is particularly useful for detailed explanations or creative outputs.
- Reduced Model Contradictions: Improvements in coherence ensure that the model's intermediate reasoning aligns more closely with its final answers.
We are rolling out a new Gemini 2.0 Flash Thinking update:
— Logan Kilpatrick (@OfficialLoganK) January 21, 2025
- Exp-01-21 variant in AI Studio and API for free
- 1 million token context window
- Native code execution support
- Longer output token generation
- Less frequent model contradictions
Try it https://t.co/fBrh6UGKz7
The Company Behind It
Google has positioned Gemini 2.0 Flash Thinking as a leader in reasoning models. The update builds on Google's long history of AI innovation, including projects like AlphaGo and earlier Gemini models. This release reflects Google's focus on advancing AI transparency and reasoning capabilities while addressing challenges such as consistency and accuracy.

Performance and Benchmarks
The updated model has demonstrated significant gains in benchmarks:
- AIME2024 (Math): Achieved a score of 73.3%, up from earlier versions.
- GPQA Diamond (Science): Scored 74.2%, showcasing its improved ability to handle scientific reasoning.
- MMMU (Multimodal Reasoning): Reached 75.4%, highlighting its strength in processing diverse data types.
These improvements place Gemini 2.0 Flash Thinking at the forefront of AI reasoning models, surpassing competitors like OpenAI's GPT-4o and Anthropic's Claude 3.5 Sonnet in certain areas.
Access and Availability
The new features are accessible for free during the experimental phase via Google AI Studio and API integration. Developers can enable code execution through the platform’s sidebar settings.