google / gemini-3.5-flash
gemini-3.5-flash
Google’s Gemini 3.5 Flash is a fast multimodal reasoning model built for agents, coding, and long-context analysis. It takes text plus media and returns text.
Overview
Gemini 3.5 Flash is a natively multimodal reasoning model from Google DeepMind. It reads text and can also interpret images, audio, and video, then produces a text response. It supports “thinking levels” that change how deeply it reasons before it answers. This makes it a strong pick for agent-style workflows where the model must plan, iterate, and keep context across many steps.
What you can build
- Coding copilots that draft, refactor, and explain code with multi-step plans
- Agent workflows that break large tasks into steps and track progress across turns
- Screenshot and chart explainers for UI debugging, dashboards, and scientific figures
- Audio summarizers for meetings, interviews, and long recordings
- Video understanding tools that describe scenes, events, and changes over time
- High-volume text pipelines for classification, extraction, rewriting, and translation
Inputs
- A required instruction written as plain text, such as a question or task description
- Optional images you want the model to interpret, with support for multiple images per request
- Optional videos for analysis, with support for multiple videos per request and long durations
- Optional audio for analysis, including long recordings
- Optional system-style instructions that set the role, tone, and rules for the whole session
- Optional reasoning depth selection (from minimal to high) to trade speed for deeper reasoning
- Optional randomness controls for generation to make outputs more deterministic or more varied
- An optional cap on response length, up to 65,536 tokens
- Optional user and session identifiers to keep multi-turn chat history consistent
Outputs
Gemini 3.5 Flash returns text. The text can be short answers, long reports, step-by-step plans, or code blocks. It can produce very long responses, up to 65,536 tokens, when you allow it.
Limitations
- The model’s output is text only. It does not generate images, audio, or video.
- The maximum input context is up to 1,048,576 tokens, and the maximum output is 65,536 tokens.
- Very long or messy inputs can still cause missed details. Clear structure helps a lot.
- Blurry screenshots, tiny on-image text, noisy audio, and low-detail video reduce accuracy.
- The model can plan actions, but it cannot take real actions without tools your app provides.
Safety & compliance
Gemini 3.5 Flash was developed under Google’s Frontier Safety Framework. It includes strengthened safeguards for cyber and CBRN-related harmful requests. Expect refusals for disallowed content, plus filtering for unsafe instructions. You are responsible for complying with applicable laws, product policies, and the rights you have to any content you upload.
API quick start
Run gemini-3.5-flash with a single API call.
{
"prompt": "Tell me about Türkiye",
"inputAll": "https://your-cdn.com/input.png",
"user_id": "...",
"session_id": "..."
}