Fast, low-cost Gemini Flash models are now live on Autohive

Gemini 3.6 Flash and Gemini 3.5 Flash-Lite are live on Autohive now. Find both under the Gemini provider in the model selector, whether you’re connecting through Gemini directly or via Vertex AI.

Neither is a frontier model. Frontier models are the biggest, most capable models on the market right now, like GPT-4o or Claude Opus, built for the hardest reasoning tasks. They also cost the most to run. Gemini Flash and Flash-Lite skip that price tag. They’re fast, affordable, and built for the kind of work most agents handle day to day.

What’s the difference between the two?

Gemini 3.6 Flash is the stronger of the pair. It handles coding assistance, document analysis, multimodal tasks, and production workloads without trouble. If your agent needs to process images, work through long documents, call tools, or stream responses in real time, this is the model to reach for. It’s priced below the frontier tier, so you get real capability without paying for reasoning headroom you don’t need.

Gemini 3.5 Flash-Lite is faster and cheaper still. It’s built for high-volume, low-latency work: classification, extraction, summarization, document processing, and background jobs. If your agent fires repeatedly, processes batches, or sits in a subagent role inside a larger pipeline, Flash-Lite earns its keep. The savings add up at scale.

On Autohive, both models list the same specs:

• Context window: 1,048,576 tokens
• Max output: 65,536 tokens
• Vision/images: yes
• Tool use: yes
• Reasoning: yes
• Streaming: yes

When to use each one

Use Gemini 3.6 Flash when your agent needs more than basic text work: document analysis, code review, multimodal inputs, or any task where you want a capable model without frontier pricing.

Use Gemini 3.5 Flash-Lite when speed and cost are the main constraints: high-volume pipelines, scheduled recurring jobs, subagent roles, or anything that runs often enough that per-call cost starts to matter.

Neither model is the right pick for tasks that genuinely need deep reasoning or high-stakes accuracy. For those, a frontier model like Claude Sonnet or GPT-4o makes more sense. Flash-tier models give up some of that ceiling for speed and throughput. For most day-to-day agent work, that’s the right trade.

Where these models fit in a multi-agent setup

One of the more practical uses for Flash-Lite is in multi-agent workflows. If you have a primary agent doing the heavy lifting, assign Flash-Lite to worker or subagent roles that handle the repetitive steps: pulling data, classifying inputs, summarizing chunks, formatting outputs. The primary agent stays on a stronger model. The workers run cheap and fast. Our multi-agent guide walks through setting this up.

Flash-Lite also pairs well with scheduled AI agents. If you’re running an agent on a daily or hourly schedule to process reports, monitor feeds, or generate summaries, the cost of running a frontier model on every trigger adds up. Flash-Lite handles those jobs without the overhead. See the automate jobs guide for setting up recurring work, and the scheduling guide for the schedule mechanics.

How to switch models

Both models are live now in the model selector when you create or configure an agent. Look for gemini-3.6-flash and gemini-3.5-flash-lite under the Gemini provider. Building a new agent from scratch? The creating your first agent guide covers the full setup, model selection included. Working from an existing agent instead? The custom agent guide covers changing models and adjusting configuration.