Generative AI Platforms: 5 Types and When to Use Each
October 24, 2026
"Generative AI platforms" sounds like one category. It's really five different layers, and most confusion comes from comparing tools that sit on different layers. Claude, Amazon Bedrock and LangChain all show up in "best generative AI tools" lists, but one is a model, one is a cloud service that hosts models, and one is a code library that connects them.
Here's the short version:
- Model provider APIs (Anthropic, OpenAI, Google Gemini API, Mistral): call one company's models directly. The fastest way to build.
- Cloud AI platforms (Amazon Bedrock, Google's Gemini Enterprise Agent Platform, Microsoft Foundry): many vendors' models inside the cloud account your company already uses.
- Open-model hubs (Hugging Face): find, test and host open models shared by the community.
- Developer frameworks (LangChain and LangGraph, LlamaIndex): code libraries for chaining models, tools and your data into apps and agents.
- End-user apps and no-code builders (ChatGPT, Claude, Gemini, n8n, Zapier Agents): use generative AI without writing code.
Pick the layer first, then the vendor. The rest of this guide explains each layer, how it charges and who it's for.
Quick Comparison
| Platform | Type | What you get | Pricing model | Best for |
|---|---|---|---|---|
| Anthropic (Claude Platform) | Model API | Claude models via the Messages API | Per token | Building on Claude directly |
| OpenAI API | Model API | OpenAI models via the Responses API | Per token | Building on GPT models directly |
| Google Gemini API | Model API | Gemini models, plus Google AI Studio | Free tier, then per token | Prototyping on a budget |
| Mistral | Model API, open weights | API access plus self-hostable models | Per token, licensing for self-hosting | Teams that may want to self-host |
| Amazon Bedrock | Cloud platform | Hundreds of models from many providers, plus agent tooling | Per token, by model | Companies already on AWS |
| Gemini Enterprise Agent Platform | Cloud platform | Gemini, third-party and open models, plus agent tools | Pay per use | Companies already on Google Cloud |
| Microsoft Foundry | Cloud platform | Large model catalog plus Foundry Agent Service | Consumption-based | Companies already on Azure |
| Hugging Face | Open-model hub | Hub of open models, Spaces, Inference Providers | Free Hub, pay-as-you-go inference | Open models and experimentation |
| LangChain / LangGraph | Framework | Open-source agent libraries, plus LangSmith | Libraries free; LangSmith has free and paid plans | Developers building agents |
| LlamaIndex | Framework | Open-source framework for data-aware agents | Framework free; cloud parsing paid | Apps built on your documents |
| ChatGPT, Claude, Gemini apps | End-user app | Chat, files, research, writing | Subscription plans; see each vendor | Individuals and teams |
| n8n, Zapier Agents | No-code builder | AI steps inside automated workflows | Plan-based | Automating work without code |
Model Provider APIs: Go Straight to the Source
If you're building a product feature, such as a support reply drafter or a document summarizer, the simplest path is one model company's API. You get the newest models first, the clearest docs and one bill.
The trade-off is lock-in. Your prompts, tool definitions and evaluations get tuned to one vendor's behavior. Switching later is possible, but it's real work.
Anthropic
Anthropic's Claude Platform gives you Claude models through the Messages API. Pricing is per million tokens, with separate input and output rates, and the page advertises discounts for batch processing and prompt caching. Anthropic also says Claude is available inside Amazon Web Services, Google Cloud and Microsoft Foundry, which matters if your company buys everything through one cloud.
OpenAI
OpenAI's quickstart describes the API as "a consistent interface to state-of-the-art AI models." The first request uses the Responses API, and official SDKs cover JavaScript and TypeScript, Python, .NET, Ruby, and Java and Go in beta.
Google Gemini API
The Gemini API is an easy place to start without paying. Google's Gemini API pricing page lists a free tier "for developers and small projects," covering input and output tokens on selected models. It also says Google AI Studio is free to use in all available regions.
Read the fine print, though: the same page says content on the free tier is used to improve Google's products, while paid-tier content isn't. Don't send customer data through the free tier.
Mistral
Mistral sells API access through Mistral Studio, billed mostly per token. The difference is that some of its models are open-weight. Mistral's pricing page says you can self-host its models and names open-weight models like Mistral 7B under Apache 2.0. It also sets separate license terms for commercial deployments, so read them before you ship.
Cloud AI Platforms: Many Models, One Cloud Account
Cloud platforms make sense once generative AI moves from a prototype to something your security team has to approve. You get many vendors' models under the identity, logging, networking and billing you already use.
Amazon Bedrock
Amazon describes Bedrock as a platform for building generative AI applications and agents at production scale, with access to hundreds of foundation models. Its pricing page lists models from Anthropic, Meta, Mistral AI, Amazon, OpenAI, Cohere, DeepSeek, Qwen, Stability AI and others, each priced per million tokens. It also offers batch inference at a lower rate than on-demand for selected models.
For agents, Bedrock AgentCore handles the plumbing of running agents securely at scale, with tracing, debugging and evaluation.
Google Gemini Enterprise Agent Platform (formerly Vertex AI)
If you last looked at Google Cloud's AI tools a year ago, the name has changed. Google's release notes say Vertex AI is now part of Gemini Enterprise Agent Platform, and Agent Builder was folded in too. Agent Engine, for example, is now called Agent Runtime.
The product page, which shortens the name to "Gemini platform," lists Google's Gemini models, third-party models like Anthropic's Claude, and open models like Gemma in Model Garden, plus tuning and evaluation tools. It's pay per use, and the page says new Google Cloud customers get a welcome credit to try it.
Microsoft Foundry
Microsoft's platform is now called Microsoft Foundry. Microsoft's Foundry documentation lists Azure AI Studio and Azure AI Foundry as its previous brand names, so you'll still see both in older tutorials. The product page describes a catalog of over 11,000 models and a built-in Foundry Agent Service. Pricing is consumption-based, and building agents needs an Azure subscription.
Open-Model Hubs: Hugging Face
Hugging Face is where open models live. The Hub hosts models, datasets and Spaces, small demo apps you can run in the browser. If you want to compare a dozen open-weight models before committing, this is where you do it.
To call those models without running your own servers, Hugging Face's Inference Providers route requests to partners like Groq, Together, Replicate, Fireworks and Cerebras through one API that works as a drop-in replacement for OpenAI's chat completions. It's pay-as-you-go, with monthly included usage on the PRO, Team and Enterprise plans. For your own always-on deployment, Inference Endpoints are billed by the hour.
Best for: teams that want open models, want to compare many options, or need to keep a self-hosting path open.
Developer Frameworks: The Glue Between Models and Your Data
Frameworks aren't model providers. They're open-source libraries that let your code chain model calls, tools, memory and data sources into an app or agent, and most of them work with any model from the layers above.
LangChain and LangGraph
LangChain is the quick-start layer for building agents with any model provider. LangGraph is its lower-level orchestration framework for agents that need reliable multi-step control, memory and human-in-the-loop approval steps. LangChain's FAQ states that LangGraph is MIT-licensed and free to use. LangSmith, the company's agent engineering platform, handles debugging, evaluation and deployment, with a free plan for solo developers and paid plans for teams.
LlamaIndex
LlamaIndex describes its framework as "an open-source framework to build agentic applications," under the MIT license. It's strongest when your app runs on your own documents. The company's commercial focus has shifted to LlamaParse, its document-parsing platform.
Choosing between the two is mostly about whether your app is document-heavy or workflow-heavy. Our LangChain vs LlamaIndex comparison goes deeper.
Generative AI Tools for End Users and No-Code Builders
Most people searching for "generative AI tools" don't want an API key. They want to write, research, summarize or automate something today.
Chat apps. ChatGPT, Claude and Gemini are the general-purpose generative AI programs most people start with. They handle writing, analysis, file uploads and research. Plans and usage limits vary, so check each vendor's plan page. The differences come down to model quality for your task, file and usage limits, and which other tools they connect to.
No-code workflow builders. When you want AI to run as a step in a process, a workflow tool fits better than a chat window. n8n lets you build AI agents inside workflows with 600+ native integrations, human approval steps and the option to self-host. Zapier Agents are AI helpers that perform tasks across 9,000+ apps.
If you're still deciding where generative AI fits in your work, start with the job, not the tool. Our guide to generative AI use cases sorts tasks by how much they tolerate a wrong answer.
Free Generative AI Tools: What's Actually Free
Plenty of these platforms have a free entry point. Here's what each one actually covers, according to the vendor:
| Platform | What's free | The catch |
|---|---|---|
| Google Gemini API | A free tier on selected models; Google AI Studio is free | Free-tier content is used to improve Google's products |
| Gemini Enterprise Agent Platform | Welcome credit for new Google Cloud customers, plus monthly free allowances for agent runtime | Credit expires; usage beyond that is billed |
| Hugging Face | Basic CPU Spaces hardware | Inference beyond included usage is pay-as-you-go |
| LangGraph, LlamaIndex | The open-source libraries (MIT) | You still pay for the models they call |
| Mistral open-weight models | Self-hosting, under Mistral's license terms | You pay for your own hardware |
A useful rule: frameworks are free, models are not. An open-source library costs nothing to install, but every model call it makes is billed by whoever runs the model, unless you host an open model yourself.
How to Choose a Generative AI Platform
| If you are… | Start with | Why |
|---|---|---|
| A developer prototyping a feature | A model provider API, or the Gemini API's free tier | Fewest moving parts |
| A company with security and compliance rules | Your existing cloud: Bedrock, Gemini Enterprise Agent Platform or Foundry | Same identity, logging and billing |
| A team that wants open models or self-hosting | Hugging Face, Mistral open weights | Control over where models run |
| Building a multi-step agent or RAG app | LangGraph or LlamaIndex on top of any model | Orchestration and data handling you'd otherwise write yourself |
| A non-developer automating work | n8n or Zapier Agents | AI inside workflows, no code |
| Someone who just wants help writing or researching | A chat app | Nothing to set up |
One caution applies to every layer: generative models produce plausible output, not guaranteed-correct output. Some problems, like fraud scoring, forecasting and ranking, are better served by classic predictive models. Our explainer on non-generative AI covers where those still win.
FAQ
What is a generative AI platform?
A generative AI platform is a service or toolkit for building with or using models that create text, images, code or audio. It covers model APIs like Anthropic's and OpenAI's, cloud platforms like Amazon Bedrock, model hubs like Hugging Face, frameworks like LangChain, and apps like ChatGPT.
What are the best generative AI tools in 2026?
It depends on the layer you need. For building, start with a model provider API or your cloud's platform. For open models, use Hugging Face. For agents, use LangGraph or LlamaIndex. For everyday work, a chat app or a no-code workflow tool is enough.
Are there free generative AI tools?
Yes. The Gemini API has a free tier, Google AI Studio is free, basic CPU Spaces on Hugging Face are free, and LangGraph and LlamaIndex are MIT-licensed open source.
What happened to Vertex AI and Azure AI Foundry?
Both were renamed. Google's release notes say Vertex AI is now part of Gemini Enterprise Agent Platform. Microsoft's documentation lists Azure AI Studio and Azure AI Foundry as previous names for what is now Microsoft Foundry.
What's the difference between a generative AI platform and a framework?
A platform runs or serves the models. A framework is code you install to connect those models to tools, data and multi-step logic. You usually need both: LangGraph, for example, calls a model hosted on Bedrock or a provider API.
Can I use generative AI programs without coding?
Yes. Chat apps need no setup, and no-code builders like n8n and Zapier Agents add AI steps to workflows through a visual editor.
Key Takeaways
- "Generative AI platforms" covers five layers: model APIs, cloud platforms, open-model hubs, frameworks and end-user apps.
- Model APIs are fastest to start. Cloud platforms win once security and procurement get involved.
- Two big names changed: Vertex AI is now part of Gemini Enterprise Agent Platform, and Azure's platform is now Microsoft Foundry.
- Frameworks like LangGraph and LlamaIndex are free and open source, but the models they call aren't.
- Free entry points exist on most layers. Check what happens to your data on free tiers.
If you'd rather run generative AI tools than assemble them from these layers, Taku is an AI-native desktop workspace where you can mirror an AI app or workflow someone already got working and keep it in your own workspace. You can browse what's there in the marketplace. Taku is in Beta, and the Mac app is available now.