Mastra launches Gateway for access to more than 500 AI models

Mastra has launched Mastra Gateway, a platform that gives developers access to more than 500 language, voice, video and decision models through a single API key and bill. The service also adds provider failover, usage controls and observability to model calls.
- Mastra Gateway provides access to more than 500 models through one API key.
- Supported workloads include language, voice, video and decision models.
- Keys can include spending limits, model permissions and US or EU regional limits.
- Traffic can fail over between multiple inference providers.
- Mastra will charge no credit card fees on inference spending until December 31, 2026.
One key for models from multiple providers
Mastra Gateway is built on the company's existing model router, which Mastra says has helped teams scale AI applications over the last two years and now runs tens of trillions of tokens every week. Unlike the router's previous setup, where each provider required a separate key and bill, the hosted Gateway makes providers available through one key on the Mastra platform.
The service includes models from Anthropic, OpenAI, Typesafe AI and other providers. According to Mastra, developers can use frontier and open source models, along with a collection of free models, across language, voice, video and decision workloads.
To get started, users create a Gateway key, add it to their project's .env file as MASTRA_GATEWAY_API_KEY, and add the mastra/ prefix to a model ID. Mastra directs users to its Gateway documentation for setup instructions, model strings and Observational Memory.
Controls for spending, models and regions
Gateway keys can be configured with hourly, daily, weekly, monthly or lifetime spending limits. Administrators can also choose which models a key is allowed to call, giving teams a way to limit access for particular agents and projects.
Mastra also provides regional settings between the US and EU. Users can set a regional limit when creating a key, or route traffic directly to a dedicated EU endpoint so prompts and user data remain within the specified region, according to the company.
The Gateway is embedded in the Mastra platform alongside deploys, databases and observability. Mastra says users can review traces for each inference call, compare models through experiments, and track performance and costs across an AI application. Platform credits with auto-top-up are also available for agent workloads.
Built-in failover and launch offer
Each model is backed by multiple inference providers. If a primary route fails or underperforms for a specific model, Mastra says the Gateway will automatically retry through another provider. The company describes this multi-provider routing as a default feature intended to keep requests served during provider disruptions.
For the launch, Mastra says it will charge no credit card fees on inference spending until December 31, 2026. The announcement positions Gateway as a way to bring model access, spending and usage into the same place as the rest of a Mastra project, while reducing the provider-specific account and routing work developers otherwise need to manage.
Mastra: Announcing Mastra Gateway | Mastra Blog


