Gemini 2.0 Flash Thinking: Google's Reasoning Model Arrives

What Flash Thinking Is

Google released Gemini 2.0 Flash Thinking in December 2025 as a specialised reasoning variant of Gemini 2.0 Flash. Like OpenAI's o1 and Anthropic's extended thinking mode in Claude 3.7 Sonnet, Flash Thinking performed extended chain-of-thought reasoning before generating its final response — spending anywhere from 10 seconds to several minutes working through complex problems before presenting an answer.

Thinking Token Visibility

Flash Thinking provided full visibility into its chain of thought — the thinking tokens were streamed to the API response in a dedicated thinking section before the final answer. Developers could display this thinking to users (as a reasoning transparency feature) or process it internally to extract intermediate reasoning steps for complex multi-stage tasks. The full thinking chain was typically 2,000-15,000 tokens for complex problems.

Benchmark Performance

On AIME 2024, Flash Thinking scored 73.7% — below o3's 96.7% but significantly above GPT-4o's 9.3% and competitive with OpenAI's o1 Mini. On MATH-500, it scored 94.4%. On GPQA Diamond, 80.3%. For many practical use cases requiring extended reasoning, Flash Thinking provided the quality improvement of dedicated reasoning models at Flash-tier pricing ($0.075/million input tokens for thinking tokens, $0.30/million for output).

Practical Applications

Google highlighted applications including: complex mathematical problem solving (useful in edtech), scientific literature interpretation, multi-step legal analysis, technical debugging requiring systematic hypothesis elimination, and financial scenario modelling. The combination of Flash-tier cost and o1-class reasoning quality made Flash Thinking the most cost-effective reasoning model available at this price point.

What This Means for Indian Businesses