On January 21, 2025, Google released Gemini 2.0 Flash Thinking—their answer to OpenAI's o1 reasoning model.

The key difference? Google made it free.

While OpenAI charged $20-$200/month for o1 access, Google gave everyone extended reasoning at no cost.

What It Offered

Extended thinking: Visible reasoning process before answering 1 million token context: Up from 32K in previous versions Native code execution: Run Python directly in responses Strong benchmarks: 73.3% on AIME math, 74.2% on GPQA science Free access: No subscription required Experimental mode: Available in Google AI Studio

Flash Thinking combined speed with deeper reasoning capabilities.

How It Worked

Like o1, Flash Thinking spent time "thinking" before responding:

• Internal reasoning: Worked through problems step-by-step
• Verification: Checked its own logic for errors
• Backtracking: Reconsidered when hitting dead ends
• Transparent process: Showed reasoning steps (unlike o1's hidden thoughts)

You could watch it think in real-time.

The Benchmark Battle

Flash Thinking competed directly with o1:

Math (AIME):

• Gemini Flash Thinking: 73.3%
• OpenAI o1: ~79%

Science (GPQA Diamond):

• Gemini Flash Thinking: 74.2%
• OpenAI o1: ~78%

Coding (Codeforces):

• Both performed at competitive programming level

Close enough to matter, especially considering the price difference.

The Strategic Play

Google wasn't trying to beat o1 on capability. They were making reasoning accessible:

• Developer adoption: Free models get more usage
• Ecosystem growth: More apps built on Google's stack
• Data advantage: More usage = more training signals
• Competitive pressure: Force OpenAI to lower prices

The long game was market share, not profit margins.

The Code Execution Advantage

Flash Thinking could run code directly:

• Solve complex calculations
• Generate data visualizations
• Test algorithms in real-time
• Debug programming problems

This native capability set it apart from text-only reasoning models.

The Context Window Win

1 million tokens meant:

• Entire codebases as context
• Full academic papers analyzed
• Long-form content generation
• Deep research synthesis

You could give it massive amounts of information to reason over.

Where Are They Now?

Flash Thinking became Google's answer to reasoning models—not the most capable, but accessible and practical. The free access attracted developers who couldn't justify o1's cost.

By late January 2025, it was clear: reasoning models were becoming table stakes, and Google was commoditizing them faster than OpenAI wanted.

January 21, 2025 was when advanced AI reasoning became available to everyone, not just those who could afford premium subscriptions—shifting the competitive dynamics of the entire AI industry.