Models, Pricing, and Limits

The gateway supports models from multiple providers through one endpoint.

Choosing a Model

Select models based on:

• Quality requirements
• Latency tolerance
• Cost sensitivity
• Feature support (vision, tool use, long context, and more)

Not all models support the same capabilities.

Current and Legacy Versions

By default, users generally retain access to:

• Current model versions
• One previous version (when available)

If you need older legacy versions for compatibility, submit a request to the support team. Legacy availability is sometimes limited by providers. For example, if a vulnerability is discovered in a legacy model, a provider may withdraw that model.

Keep this in mind when designing long-running studies.

New Model Availability

The target turnaround for onboarding newly released models is typically 24-48 hours after provider release.

Some releases may take longer because of validation, policy review, or provider capacity constraints.

Managing Spend

Use layered controls:

• Team budget caps
• Key-level limits where needed
• Rate limiting for burst control
• Routine usage reviews and cleanup of unused keys

Practical Cost Controls

1. Start with lower-cost models during development.
2. Limit max output tokens for automated workflows.
3. Cache repeatable prompts where possible.
4. Move high-value workloads to stronger models only when justified.