Gemini 3.5 Flash on Mac: What the cloud model does — and what runs locally

Gemini 3.5 Flash is Google’s stable mid-size cloud model for the Gemini API: fast, long-context, multimodal and deeply integrated into Google’s agent tooling.¹² This guide covers what it can do, how to use it from a Mac, what it costs, how Google handles your data, and where local open-weight models fit alongside it.

The short version of the local question: Gemini 3.5 Flash does not run locally. It is a Google cloud/API model with no documented weight or runtime path for Ollama, LM Studio or MLX. That does not make it useless on a Mac — most Mac users interact with cloud models daily — but it decides where your data goes and what you pay for. The details are below.

Related guides: Gemini 3.5 Flash API deep dive, Ollama on Mac mini M4, LM Studio vs Ollama, Unified Memory on Mac, Gemma 3 on Mac and the open-weight shortlist.

The graphic compares local Ollama models, Ollama Cloud, and the Gemini API by execution location, internet requirement, and data flow. It is based on the Google and Ollama documentation checked on August 14, 2026.¹³⁴

What Gemini 3.5 Flash actually is

Gemini 3.5 Flash is a stable Google model for the Gemini API and Google’s own product ecosystem. The official model ID is gemini-3.5-flash.¹²

According to Google’s model page, it accepts text, image, video, audio and PDF input and returns text output. The input context window is 1,048,576 tokens with an output limit of 65,536 tokens.¹

Feature-wise it supports thinking, function calling, code execution, Computer Use (Preview), File Search, URL context, Search Grounding, Maps Grounding, the Batch API, caching, Flex inference and Priority inference. Audio generation, image generation and the Live API are not supported.¹

The Interactions API has been generally available since June 2026 and is recommended for new projects; generateContent remains supported. Interactions are stored by default unless you set store=false.²⁵

How to use it from a Mac

There is nothing Mac-specific to install. You create an API key in Google AI Studio and call the model from any HTTP client, SDK or app that speaks the Gemini API:

Model ID: gemini-3.5-flash

In practice, three access patterns matter for Mac users:

• Scripts and apps: call the REST endpoint or an official SDK with your key. The Interactions API is the recommended surface for new projects; generateContent keeps working.²⁵
• Agent tools: coding agents and multi-model routers commonly expose the model under provider naming such as google/gemini-3.5-flash. That prefix is routing metadata, not a different model.⁴
• Google surfaces: Gemini 3.5 Flash also runs inside Google’s own products, independent of anything installed on your Mac.²

If you want the full request/response mechanics, token accounting and code samples, see the dedicated Gemini 3.5 Flash API article.

What it costs

Gemini 3.5 Flash is not “free no matter how much you use it”. Free-tier access exists, but production usage needs cost control across input, output, thinking tokens, caching, storage and grounding. Price snapshot August 14, 2026:⁶

Tier	Input	Output	Notes
Standard (paid)	$1.50 / 1M tokens	$9.00 / 1M tokens	baseline
Batch / Flex	cheaper than Standard	cheaper than Standard	latency trade-offs
Priority	more expensive	more expensive	lower latency
Grounding	per-query costs possible	—	Search and Maps Grounding

Cache writes and hits follow their own price rules on top of these numbers.⁶

Privacy: what happens to your data

This is where cloud vs. local gets concrete:⁶⁵⁷⁸

• Free-tier content may be used by Google to improve products; the paid-tier pricing table says “No”.
• Billing-enabled logs are not used for product improvement by default; projects can configure 7-, 14-, 28- or 55-day retention.⁷
• Abuse monitoring can retain prompts and outputs for 55 days, solely for policy enforcement.⁸
• Interactions are stored by default; paid-tier projects can configure 7-, 14-, 28- or 55-day retention, and store=false is available for compatible stateless requests.⁵
• Ollama says it does not see prompts or data for locally run models. For its cloud-hosted models, Ollama processes prompts and responses to provide the service and says it does not store, log or train on that content.³⁹
• For sensitive data, local-only processing or a clear enterprise/compliance workflow remains the safer default.

Does Gemini 3.5 Flash run locally in Ollama, LM Studio or MLX?

No. The official sources reviewed here contain no Gemini 3.5 Flash weights, GGUF or MLX path, and no Apple-Silicon runtime. A command like ollama run gemini-3.5-flash does not start this model on your Mac.¹⁴¹⁰

Two related traps are worth knowing:

• Provider naming is not a download. google/gemini-3.5-flash in routers or benchmark tools means provider Google, model Gemini 3.5 Flash — it is not an Ollama or local weight path.⁴
• Ollama Cloud is not local inference. Ollama documents Cloud Models that run in Ollama’s infrastructure and require an account. Useful when your Mac lacks memory, but your data leaves the machine.³

Rule of thumb: an Ollama command does not automatically mean local inference. With cloud models, Ollama acts as a local interface while inference runs elsewhere.

This is a time-bound documentation statement, not proof about every possible community project. If you see a Gemini-style entry in a local tool, check whether it is a different model, a preview/provider/cloud entry, or something that sends your prompt to a remote service.¹⁴

Local alternatives when you need offline AI

If what you want is “this class of capability, but running on my Mac”, there is no 1:1 replacement — you pick an open-weight model that fits your hardware:¹¹¹²¹⁰

Goal	Local direction	Note
Google-adjacent open weights	Gemma 3	Not Gemini, but Google’s own open family
General chat	Qwen / Llama / Mistral class	Depends on size and quantization
Coding	Qwen / DeepSeek / code-focused models	Quality varies strongly with size
Vision	Gemma 3 4B/12B/27B or vision models	Not every local model reads images
Transcription	Whisper	Different family, very practical locally
Private documents	Local RAG + Ollama/LM Studio	Privacy benefit only in a local-only setup

Rough Mac RAM orientation — heuristics, not benchmarks:¹³

• 8 GB: small 1B–4B models, sometimes heavily quantized 7B.
• 16 GB: 7B/8B comfortably, some 12B workflows.
• 24 GB: 12B/14B realistic, larger models carefully.
• 32 GB: 27B-class becomes realistic with limited context.
• 48 GB+: larger models and vision workflows get comfortable.

Runtime, quantization, context length, KV cache, swap and open apps change real memory use. For the details, see how much RAM local LLMs need.

Cloud model, local model — or both?

A practical split for most Mac users is hybrid rather than either/or:

Reach for Gemini 3.5 Flash when you need very long context, large PDF/audio/video/image analysis, function calling, code execution, File Search, Search or Maps Grounding, multi-step agents, or API integration matters more than offline privacy.¹

Reach for local models when private files should not leave the Mac, you need offline work, you want no ongoing token costs, you are building local RAG, or you process customer data, unpublished code or confidential notes.

Used together, each covers the other’s weakness: private work stays local, while the Gemini API handles the cases where its context window and tooling justify cloud processing.

Common mistakes

• Trying to install Gemini 3.5 Flash in Ollama. Correction: no official local path is documented in the sources reviewed.¹⁴
• Confusing Ollama Cloud with local inference. Correction: a cloud model means cloud processing.³
• Treating Gemma and Gemini as the same thing. Correction: Gemma is open-weight; Gemini 3.5 Flash is cloud/API.¹¹
• Using google/gemini-3.5-flash as the official Google API model ID. Correction: in the Gemini API it is gemini-3.5-flash.¹
• Assuming local AI is automatically 100% private. Correction: only with a local-only setup, no cloud tools and no exposed server.

Sources

Footnotes

1. https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰

2. https://ai.google.dev/gemini-api/docs/whats-new-gemini-3.5 ↩ ↩² ↩³ ↩⁴ ↩⁵

3. https://docs.ollama.com/cloud ↩ ↩² ↩³ ↩⁴

4. https://ollama.com/search?q=gemini ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

5. https://ai.google.dev/gemini-api/docs/interactions ↩ ↩² ↩³ ↩⁴

6. https://ai.google.dev/gemini-api/docs/pricing ↩ ↩² ↩³

7. https://ai.google.dev/gemini-api/docs/logs-policy ↩ ↩²

8. https://ai.google.dev/gemini-api/docs/usage-policies ↩ ↩²

9. https://docs.ollama.com/faq ↩

10. https://github.com/ml-explore/mlx ↩ ↩²

11. https://ollama.com/library/gemma3 ↩ ↩²

12. https://www.lmstudio.ai/docs/app/offline ↩

13. Derived from public model card sizes and community reports; ai-on-mac ran no own benchmark here. ↩