Alphabet’s Google introduced updates to its Gemini family of large language models, including a new product line designed to compete with budget-friendly AI models, such as those from Chinese rival DeepSeek.
Google provides multiple versions of its Gemini models, each differing in price and performance. While it previously offered a lightweight variant called “Flash,” the newly introduced “Flash-Lite” is even more cost-effective. On Wednesday, the company rolled out Gemini 2.0 Flash to the public after a developer preview in December. Additionally, it launched Flash-Lite and began testing an updated version of its flagship “Pro” model.
According to Koray Kavukcuoglu, Chief Technology Officer of Google’s DeepMind AI lab, the company developed Flash-Lite after receiving positive feedback on the 1.5 version of Flash. In a press release, he noted that Gemini 2.0 Flash is more expensive than its predecessor. Recently, the cost of developing and using AI models has drawn investor attention, especially after DeepSeek disclosed that its final training run cost under $6 million. However, developers at top U.S. AI firms suggested that the actual expenses were likely significantly higher.
DeepSeek’s emergence has raised questions during earnings calls for Alphabet, Microsoft, and Meta, which have reaffirmed their commitment to significant capital investments in AI.
On Tuesday, Alphabet’s shares declined, partly due to investor concerns over a planned increase in capital expenditures, which exceeded Wall Street expectations by 29%.
Google prices certain inputs on Gemini Flash-Lite at $0.019 per 1 million tokens, the data units an AI model processes. OpenAI charges $0.075 for its cost-efficient flagship model, while DeepSeek sets its budget model at $0.014. However, DeepSeek states on its website that it will raise its pricing fivefold on February 8.
