How to Use Gemini 2.5 Pro Inside Sandbox AI

Gemini 2.5 Pro is the multimodal heavyweight: it accepts million-token contexts and natively understands video, audio and images — not as bolted-on adapters, but as first-class input. In Sandbox, its Deep Research capability runs as a delegated agent while you keep working.

What makes Gemini different

Native video input — upload an hour of footage and ask questions, get timestamps, build summaries. No transcription middleman.
Million-token context — entire codebases, full meeting archives, complete book manuscripts in a single request.
Deep Research — multi-hour autonomous web research producing cited reports; in Sandbox it runs as a background agent and pings you when done.
Imagen 4 — Google's image model is included in the same credit pool.

Gemini 2.5 Pro — flagshipmultimodal

Gemini Flash — fast lanecheapest

Deep Research agentbackground

Imagen 4 — image generationincluded

The Gemini family in your workspace

Workflows Sandbox adds on top

Because Sandbox runs models in parallel, Gemini's multimodality becomes a testing tool: feed the same product video to Gemini and Grok and Claude, compare what each one understood. Editors use it to verify that a video's message actually lands. Researchers use Deep Research agents for due diligence while GPT-5 drafts the summary template — everything lands in one workspace with one bill.

⌘[upload 47-min webinar] Extract every product claim made after 00:12:00, with timestamps, as a table.

⌘[upload repo zip] Map the dependency graph and flag the three modules most likely to break under the planned API change.

⌘Deep Research: 10-year trend analysis of EU AI regulation, with sources. Run in background.

EU note: Gemini requests route to Google's EU processing regions by default for EU customers; Ultra plans can pin the entire workspace to Frankfurt.

Why not just use the Gemini app?

Google's own app is a fine way to talk to one model. It won't tell you that GPT-5 caught something Gemini missed on the same brief, and it doesn't include Midjourney, ElevenLabs or an autonomous research agent. Install Sandbox on Windows once and Gemini's video understanding sits in the same window as every model you'd otherwise pay for separately — one bill, one login.

Using Gemini 2.5 Pro inside Sandbox

What makes Gemini different

Workflows Sandbox adds on top

Why not just use the Gemini app?

Gemini, GPT-5, Claude — same prompt, same screen

Keep reading