Gemini 3.5 Flash
Google's fast, cost-optimized agentic model with built-in computer use
Gemini 3.5 Flash is Google DeepMind's speed- and cost-optimized model in the Gemini 3.5 family, launched at Google I/O on May 19, 2026. It is built for agentic workloads, offering frontier-level reasoning and coding at a fraction of the cost of flagship models. In June 2026 Google added native 'computer use' as a built-in tool, letting developers build agents that can see a screen and click, type, scroll, and navigate across browser, mobile, and desktop environments. This capability, previously available only via the standalone Gemini 2.5 computer use model, now runs inside the same model developers use for function calling, Search grounding, and Maps integration, enabling a single agent to combine screen interaction with these tools. The model scored 78.4% on OSWorld-Verified and operates through a continuous observe-think-act loop, with each action returning a structured command (coordinate click, keystroke, scroll, or form entry). It supports text, image, video, audio, and PDF inputs, a ~1M-token context window, and up to ~65K output tokens. To address safety risks, Google applies targeted adversarial training against prompt injection and offers two optional enterprise safeguards: requiring explicit user confirmation for sensitive or irreversible actions, and automatically halting tasks when an indirect prompt injection is detected. It is available to developers and enterprises through the Gemini API and the Gemini Enterprise Agent Platform, with a live demo hosted by Browserbase.
Who it's for
Pricing · usage-based
checked 1d ago| Plan | Price | Includes |
|---|---|---|
| Free tier (Gemini API) | $0 /mo | Free access to the Gemini API with rate limits for testing and prototyping · 5,000 free Google Search grounding prompts/month across the Gemini 3.x family · Browserbase demo environment to try computer use |
| Paid API (pay-per-token) | Free | $1.50 per 1M input tokens · $9.00 per 1M output tokens · Cached input at $0.15 per 1M tokens · 50% discount on batch requests · Native computer use, function calling, Search and Maps grounding |
| Enterprise (Gemini Enterprise Agent Platform) | Free | Custom enterprise pricing and deployment · Optional safeguards: user confirmation for sensitive actions and prompt-injection halting · Support for enterprise automation like continuous software testing and knowledge work |
AI-researched pricing — verify on the official site before subscribing.
Use it for
- — Building custom AI agents that navigate browsers, mobile apps, and desktops
- — Continuous, resilient software UI testing without brittle selectors
- — Automating multi-step enterprise workflows and knowledge work
- — High-volume, low-latency coding and reasoning tasks
- — Combining screen interaction with Search and Maps grounding in a single agent
Get the most out of it
- 01Use computer use for long-horizon, interface-driven tasks where visual reasoning beats fixed selectors, but keep humans in the loop for critical steps
- 02Enable the optional enterprise safeguards (user confirmation and prompt-injection halting) since they are opt-in, not on by default
- 03Scope agent identity and permissions narrowly so a compromised agent can't delete records or approve payments
- 04Use context caching ($0.15/M cached input) and the 50% batch discount to cut costs on repetitive or large-context agentic loops
- 05Run agents in a secure sandbox with strict access controls to contain failures from misclicks or malicious content
- 06Prototype on the free Gemini API tier and the Browserbase demo before moving high-volume workloads to paid pay-per-token