AI Tools
What Does Modular MAX Cost in 2026?
Modular MAX pricing starts at $0: the Community Edition is free forever for self-hosted MAX and Mojo, while Modular Cloud charges per token for batch workloads or per GPU-hour for dedicated endpoints, and Enterprise is custom-priced. Exact rates are not published. This pricing was verified on Modular's official pricing page as of August 21, 2026.
What matters
- Mojo went fully open source on August 18, 2026 under Apache 2.0 with LLVM exceptions; the source lives in the modular/modular GitHub repository, which shows 28,362 stars as of August 21, 2026.
- Modular MAX has a $0 tier: the Community Edition is free forever for self-managed deployments of MAX and Mojo, per the official pricing page.
- Modular Cloud is usage-based: shared endpoints are free to test, batch inference is priced per token, and dedicated endpoints are priced per GPU-hour.
- Enterprise is a custom, sales-led engagement; the exact per-token and per-GPU-hour rates are not published on the pricing page.
What is Modular MAX?
Modular MAX is a unified AI inference platform from Modular: a software stack that runs generative AI models on NVIDIA and AMD GPUs, from custom GPU kernels to production cloud serving, per the official pricing page. Mojo is Modular's programming language, which the company describes as "the systems language for the AI era." MAX and Mojo together form the Modular Platform, hosted in the modular/modular GitHub repository, which shows 28,362 stars, was created in April 2023, and was pushed most recently on August 21, 2026.
The distinction matters for buyers: Mojo is the language you write in, MAX is the runtime and serving layer that executes and deploys your models. Both are part of the same open-source repository as of August 2026, and both feed into Modular's paid cloud service.
The news: Mojo is now open source (August 18, 2026)
On August 18, 2026, Modular announced at ModCon 2026, themed "Open source, open cloud, open silicon," that the Mojo language is now fully open source under the Apache 2.0 license with LLVM exceptions. The announcement covers the Mojo compiler, tooling, and everything else needed to build the language, all published in the modular GitHub repository. The repository's LICENSE file confirms the terms: "The Modular repository is licensed under the Apache License v2.0 with LLVM Exceptions."
Modular says the LLVM extensions to the license further expand the freedoms for building and distributing binaries compiled from Mojo. The company frames the release as making the language fully open source, which means the self-hosted stack is now free to run, modify, and redistribute without license fees. For anyone evaluating MAX as an inference platform, this removes the licensing question entirely and leaves only the cloud pricing decision.
Modular MAX pricing: the four editions
Modular's pricing page, checked on August 21, 2026, lists four editions. The Community Edition is $0: free forever, self-managed deployment of MAX and Mojo with community support. Modular Cloud is the managed service, described as "fully managed, pay by usage," where you start by testing shared endpoints for free before moving to dedicated endpoints. Batch inference workloads are priced per token. Dedicated Endpoints are priced per GPU-hour, with always-on compute and support for custom models. Enterprise is a custom engagement covering compliance, control, and GPU flexibility, with dedicated support and forward-deployed engineers.
One detail the pricing page does not publish: the exact per-token and per-GPU-hour rates. The page funnels you to signup or a sales conversation instead. If you need a hard number before committing, budget for a quote or an account-based pricing view rather than a public rate card.
What the free tier actually includes
The $0 tier is self-hosted, not cloud. The Community Edition gives you the software, not the GPUs: you run MAX and Mojo on your own NVIDIA or AMD hardware and manage the deployment yourself. Modular describes the Self-Hosted Community Edition as "completely free and open source."
The paid product is Modular Cloud: managed inference endpoints with usage metrics, observability, and Modular engineers optimizing your workloads. The honest framing for a buyer is: Mojo the language is free, MAX self-hosted is free, but managed cloud inference is a usage-based expense. Teams that do not want to operate GPUs will pay per token or per GPU-hour; teams with existing hardware get a genuinely free path.
Who should choose which edition
- Hobbyists, researchers, and teams with spare NVIDIA or AMD GPUs: the Community Edition at $0, self-hosted.
- Product teams that want managed API access without running infrastructure: Modular Cloud shared endpoints first, then dedicated endpoints as load grows.
- Offline or high-volume batch workloads: the Batch API, priced per token.
- Latency-sensitive production services and custom models: Dedicated Endpoints, priced per GPU-hour.
- Regulated organizations that need data to stay inside their VPC: Your Cloud (BYOC) or the Enterprise engagement, custom-priced.
Modular positions the platform as running "across them all" without locking developers or infrastructure operators to a single hardware vendor, per its pricing page FAQ. That portability, from your own hardware to a managed cloud, is the core of its pitch.
How Modular MAX compares with alternatives
Modular competes with managed inference providers such as OpenRouter, Together AI, Fireworks, and Groq. Two differences matter for buyers comparing options. First, transparency: OpenRouter's models page is titled "Compare AI Models: Pricing, Context & Benchmarks" and publishes per-model rates openly, while Modular publishes no unit prices on its site. Second, deployment freedom: few competitors let you run the same software stack in your own VPC or on your own hardware, which is Modular's stated differentiator.
If published, budgetable pricing is your priority, an aggregator like OpenRouter is easier to cost out. If you want one stack that moves from self-hosted to managed without rewriting, MAX is the argument to make. For a fuller look at model-aggregation services, see our OpenRouter alternatives guide; for the inference-pricing landscape, our DeepSeek API pricing analysis is a useful benchmark.
Limitations to know before you buy
Three gaps are worth naming before you evaluate Modular MAX against rivals. Exact unit rates are not public: per-token and per-GPU-hour prices require signup or a sales contact, so spreadsheet-level cost comparison is not possible up front. Free means self-hosted: there is no published free quota on Modular Cloud, and the $0 edition assumes you bring your own hardware. Enterprise is sales-led: no starting price is published, and the VPC and on-premise deployment options route through the Enterprise engagement rather than self-serve checkout.
At a glance
| Edition | Deployment | Pricing | Support |
|---|---|---|---|
| Community Edition | Self-managed MAX and Mojo on your own hardware | $0, free forever | Community |
| Modular Cloud shared endpoints | Managed API, pay by usage | Free to test, then usage-based | Modular |
| Batch API | Managed batch inference workloads | Per token | Modular |
| Dedicated Endpoint | Managed, always-on compute for custom models | Per GPU-hour | Forward-deployed engineers |
| Enterprise | Your VPC or on-premise (BYOC) | Custom, sales-led | Dedicated |
FAQ
Is Mojo free to use now?
Yes. Mojo is fully open source under the Apache 2.0 license with LLVM exceptions, announced August 18, 2026. The compiler and tooling source are in the modular/modular GitHub repository, and the LICENSE file confirms the Apache 2.0 with LLVM Exceptions terms.
How much does Modular MAX cost?
The Community Edition costs $0 and is free forever for self-hosted deployments. Modular Cloud charges per token for batch workloads and per GPU-hour for dedicated endpoints, with shared endpoints free to test. Enterprise is custom-priced. Exact per-unit rates are not published on the pricing page.
Is Modular MAX open source?
The self-hosted stack is. The modular/modular repository, which includes both MAX and Mojo, is licensed under Apache 2.0 with LLVM Exceptions, and Modular calls the Self-Hosted Community Edition completely free and open source. Modular Cloud, the managed service, is paid.
Related reading
OpenRouter alternatives, DeepSeek API pricing, ToolBistro radar index