Lovable Adds Gemini 3.5 Flash to Power Agentic Features Without API Setup
Lovable now lets you swap Gemini 3.5 Flash into any app you build, unlocking faster, cheaper multi-step agentic features without touching an API key.
Takeaways
- Gemini 3.5 Flash is now available in Lovable's built-in AI connector for AI features inside apps you build.
- Zero setup required: no API keys, no backend wiring needed; Lovable manages credentials and streaming automatically.
- Built for agentic workflows: Gemini 3.5 Flash scores 83.6% on MCP Atlas multi-step tool benchmarks, best among all tested models.
- Thinking levels: the model lets you tune quality vs. latency/cost with configurable reasoning depth.
- Pass-through pricing at $0.30/M input and $2.50/M output tokens, with $1 free AI usage per workspace per month.
- Separate from the Lovable build agent: this model powers features in your deployed app, not the coding assistant that builds it.
Lovable, the AI-powered app builder, has added Gemini 3.5 Flash to its built-in AI connector, making Google DeepMind's latest fast reasoning model available for the AI features you embed inside your apps. The pitch is simple: if you're building something with agentic behavior, like a multi-step workflow, a coding assistant, or a document Q&A bot, you can now route those calls through Gemini 3.5 Flash instead of the default model.
Two layers of AI, one platform
There's an important distinction worth understanding here. The default model applies to AI features inside your app. It is not the model Lovable uses to write, edit, or reason about your code. In other words, Lovable uses its own model to help you build the app, while the model you pick from the connector is what powers the AI features your end users interact with at runtime.
AI gateway usage is measured when AI features inside your deployed app make model calls. These requests are separate from the Lovable agent that helps you plan, build, and edit your project. Usage shows up as Run credits in your settings dashboard.
What Gemini 3.5 Flash actually is
Gemini 3.5 Flash is the next iteration in the Gemini 3 series of highly-capable, natively multimodal, reasoning models. It is based on the Gemini 3 Flash reasoning foundation with thinking levels to control the mix of quality, cost and latency. Think of "thinking levels" as a dial that lets you trade off how much the model reasons before responding, which directly affects both latency and cost.
Gemini 2.5 Flash breaks the classic speed-vs-quality trade-off, delivering frontier-level intelligence at latency that makes it practical for real-time applications and multi-step agentic workflows. The Gemini 3.5 generation pushes this further, with the model card benchmarks showing particularly strong results in agentic tool use.
The benchmark story
The numbers from Google DeepMind's own evaluation are worth looking at. On agentic benchmarks specifically:
- Gemini 3.5 Flash scores 83.6% on MCP Atlas (multi-step workflows using MCP), the highest score among all models tested including GPT-5.5 and Claude Opus 4.7.
- It scores 56.5% on Toolathlon (real-world general tool use), edging out GPT-5.5 at 55.6%.
- On SWE-Bench Pro (diverse agentic coding tasks), it hits 55.1%, ahead of Gemini 3.1 Pro but behind Claude Opus 4.7 at 64.3%.
- Google's own assessments confirm that Gemini 3.5 Flash excels at agents and coding.
The one area where it trails is long-context recall at 128k tokens, where GPT-5.5 scores 94.8% vs Gemini 3.5 Flash's 77.3%. For most app-level use cases that won't matter, but it's worth knowing if you're building document-heavy pipelines.
No API keys, no backend wiring
The practical appeal of this integration is the zero-setup story. Normally, adding a model like this to an app means managing API credentials, setting up a secure backend to avoid exposing keys in the browser, and handling streaming responses yourself. The built-in AI connector uses Gemini 3 Flash (preview) as its default model. When asking Lovable to add or update an AI feature, you can specify a different supported model or combination of models for that feature.
Lovable handles the infrastructure automatically:
- A
LOVABLE_API_KEYis generated and managed per project, no provider account needed. - All AI calls run through a secure backend edge function, never directly from the browser.
- Streaming is supported out of the box via server-sent events (SSE), so chatbot responses appear token-by-token.
To use Gemini 3.5 Flash specifically, you just tell Lovable which model you want when prompting it to add or update an AI feature in your app.
Where it fits in the model lineup
According to Lovable's own documentation, Gemini 3.5 Flash sits in the "best balance" tier alongside Gemini 3 Flash, GPT-5 Mini, and Gemini 2.5 Flash, for use cases where speed, cost, and intelligence all matter. The full supported model list now spans from ultra-cheap options like Gemini 3.1 Flash Lite all the way up to GPT-5.5 Pro for frontier reasoning tasks.
The sweet spot for Gemini 3.5 Flash is any feature that needs to take multiple steps, call tools, or reason through a problem before responding, without the latency or cost of a full Pro-tier model. Think:
- In-app coding assistants that can run and iterate on code
- Workflow automation that chains tool calls (search, write, summarize)
- Document Q&A with structured reasoning over retrieved content
- Agentic chatbots that need to decide between multiple actions
Pricing
Lovable passes through model costs at provider rates with no markup. Pricing for Gemini 2.5 Flash starts at $0.30 per million input tokens and $2.50 per million output tokens. Every workspace gets $1 of free AI usage per month to start, with paid plan users able to top up their balance as needed. The Gemini 3.5 Flash pricing follows the same pass-through model.
For teams already building on Lovable, this is a low-friction upgrade path. If your app has any agentic features and you haven't explicitly chosen a model, swapping the default to Gemini 3.5 Flash is a one-sentence prompt away.