Google ships the Gemini 2.0 family as a tiered lineup: Flash to GA, Pro and Flash-Lite in preview

Four models, one release, four different jobs

The headline is that Gemini 2.0 Flash is now generally available through the Gemini API in Google AI Studio and Vertex AI, meaning developers can build production applications on it rather than experimental previews. But the announcement is really about a spread of tiers going out on the same day.

Alongside the GA Flash, Google released an experimental Gemini 2.0 Pro, a public-preview Flash-Lite, and made 2.0 Flash Thinking Experimental selectable in the Gemini app's model dropdown. Each occupies a different point on the cost-latency-capability curve: Flash as the high-volume workhorse, Pro for coding and complex prompts, Flash-Lite for the cheapest usable option, and Flash Thinking for step-by-step reasoning.

The context window is the visible differentiator

The clearest technical split in this lineup is context length. Both 2.0 Flash and 2.0 Flash-Lite carry a 1 million token context window, while 2.0 Pro doubles that to 2 million tokens — described as Google's largest — and adds the ability to call tools like Google Search and code execution.

That framing positions Pro as the model for tasks where you need to load and reason over very large bodies of information at once, not just generate quick responses. All four models launch with multimodal input and text output, with Google saying additional output modalities like image generation and text-to-speech are coming in later general availability.

Flash-Lite's pitch is a price-hold, not a price cut

Flash-Lite is described as the most cost-efficient model yet, but the specific claim is subtler than 'cheaper.' Google says it delivers better quality than 1.5 Flash at the same speed and cost, and outperforms 1.5 Flash on the majority of benchmarks. In other words, the strategy is to raise the quality floor while holding the previous generation's price and latency.

The single concrete cost figure Google offers is an illustration: Flash-Lite can generate a one-line caption for roughly 40,000 unique photos for under a dollar in Google AI Studio's paid tier. That is the only pricing detail in the announcement itself — everything else is deferred to the Google for Developers blog.

A safety note that names indirect prompt injection

The safety section is worth reading closely because it names a specific mechanism. Google says the 2.0 lineup was built with reinforcement learning techniques that use Gemini itself to critique its own responses, which it credits with more accurate feedback and better handling of sensitive prompts.

We're also leveraging automated red teaming to assess safety and security risks, including those posed by risks from indirect prompt injection, a type of cybersecurity attack which involves attackers hiding malicious instructions in data that is likely to be retrieved by an AI system.Montana Labs

Naming indirect prompt injection matters given that Pro ships with tool calling and search. Once a model retrieves external data and executes tools, the attack surface described here stops being theoretical — the safety work and the agentic capabilities in this release are two sides of the same launch.

What the tiering forces developers to decide

The practical implication of this announcement is that choosing a Gemini model is now a routing decision, not a single default. Only Flash is GA and production-ready today; Pro and Flash-Lite remain experimental or in preview, so teams building now must weigh a stable workhorse against a more capable Pro they cannot yet fully rely on.

For applied teams, the sensible read is to build on GA Flash for production traffic, prototype long-context and coding-heavy workloads against experimental Pro, and evaluate Flash-Lite as a drop-in upgrade path from 1.5 Flash where cost sensitivity dominates. The four-tier structure is the product — the work is matching each task to the tier whose context window and cost profile actually fits it.