Google DeepMind logo

Gemini 3 Pro on live video streams

Gemini 3 Pro · Google DeepMind · Multimodal · Closed API

Gemini 3 Pro is Google DeepMind’s frontier reasoning tier of Gemini 3, released in November 2025 with a 1M-token context window and state-of-the-art long-video and document understanding. On Overshoot it is available as a passthrough model, reachable through the same stream publishing and OpenAI-compatible chat-completions request format used across the API.

Gemini 3 Pro is available through the Overshoot API as a passthrough model (gemini-3.1-pro-preview, status: ready, verified against the live model catalog on 2026-07-14) - requests are forwarded to the upstream provider rather than running on Overshoot's hosted infrastructure.

Developer
Google DeepMind
Parameters
Undisclosed
Context window
1M tokens
License
Proprietary
Released
Nov 2025
Inputs
Text, images, video frames
Overshoot availability
API passthroughas of 2026-07-14

What Gemini 3 Pro is good at

Gemini 3 Pro is Google DeepMind’s frontier reasoning tier, built for the hardest visual understanding tasks: long-video comprehension, dense multi-page documents, and scenes that require weighing several details against each other before answering. It carries the deepest reasoning available in the Gemini 3 family.

The 1M-token context window pairs naturally with that strength, since long-video and document understanding both benefit from keeping extensive context, many sampled frames, and prior conversation turns available in a single request.

  • State-of-the-art long-video and document understanding
  • Deep multi-step visual reasoning over complex scenes
  • A 1M-token context window that scales with long-form input

Running Gemini 3 Pro on Overshoot

Publish a camera or screen share over WebRTC through LiveKit to open a Stream, then send a chat-completions request whose image_url or video_url is an ovs:// reference to the latest frame, an exact timestamp, or a recent segment drawn from the 600-second retention window.

As a passthrough model, Gemini 3 Pro reaches the API through the same OpenAI-compatible endpoint used across Overshoot: the platform manages stream lifecycle, frame selection, and SSE streaming transport, then forwards the resolved request to Google. The current passthrough id in the live catalog is gemini-3.1-pro-preview, as returned by GET /v1beta/models.

Gemini 3 Pro versus Gemini 3 Flash

Pro pushes further than Gemini 3 Flash on reasoning depth and long-context comprehension, at higher cost and latency. Teams typically default to Flash for routine, high-volume visual queries and reserve Pro for tasks that specifically need its long-video or document reasoning ceiling.

Frequently asked questions

Can Gemini 3 Pro analyze live video?

Yes. Overshoot carries the live WebRTC stream and resolves an ovs:// reference to the latest frame or a recent segment, then forwards the chat-completions request to Google, where Gemini 3 Pro generates the response.

Is Gemini 3 Pro open source?

No. Gemini 3 Pro is a closed Google DeepMind model reachable only through Google’s API. Overshoot offers it as a passthrough option, with key management shared with its hosted open-weight models.

How fast is Gemini 3 Pro on Overshoot?

As a passthrough model, Gemini 3 Pro’s latency depends on Google’s own serving stack rather than Overshoot’s hosted path. Overshoot still handles stream publishing, frame selection, and SSE streaming transport, so the request and response shape match a hosted model.

What is Gemini 3 Pro best used for?

Gemini 3 Pro fits tasks that need deep reasoning over long video segments or dense documents, such as reviewing an extended recorded feed or extracting structured detail from many pages at once.

Related models