Hedra
Free tierInference, fine-tuning and GPU runtime for frontier visual AI models
Free tier available·All audiences·Powered by Hedra (proprietary frontier visual models) plus 100+ third-party models·API available
Key strengths
Frontier visual model research and inferenceWorld Model Runtime (WMR) purpose-built for diffusion, video, and world modelsManaged inference with 100+ modelsFine-tuning and training with reinforcement learning for task-specific objectivesGPU sandboxes with flexible scheduling and scalingHybrid, on-prem, and sovereign deployment optionsNo-code Studio for image and video creation
Free tier + paid plans · from $20 USD/mo
Self-hostable
Facts verified against hedra.com on
Recent changes
- Self-hosting: now available
- Models updated: Hedra (proprietary frontier visual models) plus 100+ third-party models
- Plans updated: Free plan available (no credit card). Paid plans: Basic $20/mo (2,000 credits), Pro $50/mo (7,200 credits), Ultra $100/mo (18,000 credits, up to 57,600 credits/month), Teams $75/mo (14,400 credits). Enterprise: custom pricing with dedicated support, private deployments, SSO, and more.
Get alerts when Hedra changes
Free. We'll email you when pricing, the free tier, API access or open-source status changes.
No ratings yet
Hedra — Developer Documentation & Integration
Hedra exposes its capabilities through a REST API, giving developers programmatic access to inference, fine-tuning, and GPU sandbox environments. Key integration points:
- API access: Available across all paid tiers; enables inference calls against 100+ models including Hedra's proprietary frontier visual models.
- Fine-tuning endpoints: Trigger and manage fine-tuning jobs with reinforcement learning for task-specific objectives directly via API.
- GPU sandboxes: Provision and manage GPU sandbox environments with flexible scheduling and scaling — suitable for both experimental and production pipelines.
- World Model Runtime (WMR): Underlying runtime is purpose-built for diffusion, video, and world model workloads, meaning inference latency and throughput characteristics differ from standard transformer serving stacks.
- Self-hosting: The platform is open-source and self-hostable, supporting hybrid, on-prem, and sovereign deployment configurations.
- Enterprise controls: SSO, private deployments, and dedicated support are available under custom enterprise agreements.
Developers should consult Hedra's official documentation for endpoint references, authentication, credit consumption rates per model call, and SDK availability.