Google brings Veo 2 and Live API to developers, teases Gemini 2.5 Flash

Developers usually have to make an annoying trade-off: pick a slow model that can reason through complex code, or pick a fast one that can handle real-time audio and video.

At this year's Cloud Next, Google decided to put all those capabilities directly into the Gemini API.

Even the Flash model is getting a brain

Just last week, Google made Gemini 2.5 Pro available to developers in Google AI Studio and Vertex AI. It is their flagship thinking model, designed to reason before answering, with a heavy focus on agentic coding and building complex web apps.

Now, Google says Gemini 2.5 Flash is coming soon.

Flash has always been the cheap, low-latency workhorse of the lineup. Google's plan is to bring adaptive thinking to every Gemini model, not just the expensive Pro tier.

With a massive one-million-token context window, a reasoning Flash model means developers can run multi-agent systems and analyze entire codebases without blowing through latency budgets or bank accounts.

Veo 2 goes GA, and Live API enters preview

The update also brings two major upgrades for media and real-time interaction.

Google's video generator, Veo 2, is now production-ready and generally available in the Gemini API. It handles both text and image prompts, follows complex camera directions, and simulates physical interactions across different styles.

One gaming studio, Wolf Games, is using Veo 2 for interactive cinematic games and reported slashing their visual iteration cycles by over 60% while drastically cutting production time.

Meanwhile, the Live API, which launched experimentally back in December, is officially moving to Preview. It processes bidirectional streaming audio, video, and text with minimal latency, making it ideal for live translation, meeting assistants, and real-time environment monitoring.

Google bumped up rate limits significantly and baked in tools like Google Search, code execution, and function calling to work seamlessly alongside Gemini 2.0 Flash.

Why it matters

For builders, this is about getting a complete multimodal stack from a single endpoint. You can now take a live video stream, feed it through a low-latency model with tool use, kick complex reasoning off to a 2.5 model, and generate video assets with Veo 2 without stitching three different providers together.

It is definitely worth testing in AI Studio right now, especially since the Live API limits are finally high enough to run serious prototypes.

For the broader industry, Google is shifting the battle from isolated model benchmarks to end-to-end multimodal infrastructure. While rivals offer video generation or advanced voice through separate consumer apps, Google is packaging everything into an API.

We think adding reasoning to Flash is the smartest move here. Pro models grab the headlines, but cheap and fast models run the actual world. If Flash can think adaptively without spiking latency, it could become the default brain for production agents. The main friction point will be managing multimodal bandwidth and total cost once streaming video and generation run at scale.