Gemini 3.8 Flash is generally available
Google's newest Flash model, now the default for most of our new builds. Older 3.5 and 3.7 Flash names now route to newer versions.
We build on Google's Gemini models to understand documents, images, audio and video together, answer from live Google Search and Maps data, and run voice and workflow agents — on AI Studio or Vertex AI.
Gemini reads all of these together and answers across them.
Illustrative demo with sample content.
Here are the releases and retirements that change what you can build, and what has to be updated.
Google's newest Flash model, now the default for most of our new builds. Older 3.5 and 3.7 Flash names now route to newer versions.
Low-latency speech in and out through the Live API, including a version with extended thinking.
Generally available for image generation and editing.
Video generation up to 4K, with clip extension and first-and-last-frame control.
Google-hosted agents in public preview, with the Antigravity agent updated in September.
Open-weight models you can run on your own servers.
temperature, top_p and top_k are deprecated for current models — prompts that relied on them need retesting.
Source: Gemini API release notes. Checked October 2026.
Gemini was built to handle every kind of input. Choose one to see what it does and which model we use.
A model on its own can sound sure and still be wrong. Grounding makes Gemini look things up first and show its sources.
Tell us your volume and needs. We match a model, then pick the inference tier that fits your latency and budget.
Google's agent tools have matured quickly. We combine them with your systems, permissions and approval steps.
Agents that see the screen and click, type and scroll in browsers and web apps — for portals that have no API.
Research agents that search, read dozens of sources and write a structured report with citations.
Google-hosted agents (public preview) that run long tasks with tools, so you do not manage the loop yourself.
One API for multi-step, multimodal conversations with built-in tools, your functions and Python execution.
Both give access to Gemini models. The right one depends on your size, security needs and how much of Google Cloud you already use.
| Gemini API (AI Studio) | Vertex AI (Google Cloud) | |
|---|---|---|
| Best for | Prototypes, start-ups, fast iteration | Enterprises and regulated workloads |
| Sign-in and billing | Google account, AI Studio billing, spend caps | Google Cloud project, IAM, existing GCP billing |
| Data and security | Paid tier: prompts not used to improve products | VPC Service Controls, CMEK, data residency, audit logs |
| Extras | Quick API keys, Interactions API, AI Studio tools | Model Garden, MLOps, BigQuery and Workspace integration |
Google retires older Gemini, Imagen and Veo models on fixed dates. We track them and upgrade our clients' apps before anything breaks.
Each one starts with a clear business result and the data it needs.
Gemini is often the best fit for multimodal, voice and location-aware apps, especially on Google Cloud. When another model suits you better, we say so.
| Model family | Strengths | When we choose it |
|---|---|---|
| Google Gemini | Multimodal (text, image, audio, video), long context, Google Search and Maps grounding | Products on Google Cloud or Workspace; video, voice and location-aware apps |
| OpenAI | Broad APIs, realtime voice, agents and images | Voice agents and multimodal products |
| Anthropic Claude | Long documents, coding and dependable agents | Contracts, research and coding assistants |
| Meta Llama | Open weights to host yourself | Strict on-premise needs |
| DeepSeek | Low-cost open reasoning | Cost-sensitive self-hosted work |
| Qwen | Open-weight, multilingual | Multilingual products |
Straight answers to what clients ask us most.
For most apps we start with Gemini 3.8 Flash. Very high-volume, simple tasks go to Flash-Lite, the hardest reasoning to Pro, and real-time voice to Gemini 3.8 Live. We confirm the choice with evals on your own data.
The Gemini API through AI Studio is the quickest way to build and suits start-ups and prototypes. Vertex AI on Google Cloud adds enterprise controls — IAM, VPC Service Controls, customer-managed keys and data residency — and is what we use for regulated or large organisations.
Gemini 2.0 Flash was shut down on 1 June 2026 and access to 2.5 models is now limited to recent users. We move apps to the 3.x family, remove deprecated sampling settings, and re-run evals so answers stay consistent.
Grounding lets Gemini check Google Search, Google Maps, your files or specific URLs before answering, and return the sources it used. It sharply reduces made-up answers for anything that changes over time, such as prices, opening hours or policies.
Yes. Gemini can watch long videos and listen to recordings in the same request as text and documents, and the Live API supports real-time voice conversations. For generation, Omni and Veo create video and Nano Banana creates images.
We route easy work to Flash-Lite, use the Batch API and the cheaper Flex tier for work that can wait, cache repeated context, keep outputs short and set project-level spend caps so bills stay predictable.
Yes. We build add-ons and agents that work with Gmail, Docs, Sheets and Drive data through Google's APIs, using your existing Google sign-in and permissions.
Tell us what you want Gemini to read, see, hear or do. We reply within 4 business hours with honest advice.