NVIDIA B300 GPU compute
GPU compute billed per GPU-second; equivalent to $8.25 per GPU-hour.
- $8.25 / GPU-hour equivalent
- Billed per GPU-second
AI Video Generator

PRODUCT OVERVIEW
Hedra is a research and infrastructure company that brings frontier inference to multimodal and physical AI. Its managed v3 API provides one shared asynchronous job lifecycle for generating images, video, and audio, plus OpenAI- and Anthropic-compatible chat endpoints for vision, reasoning, and language tasks. A model catalog spans image, video, audio, and text models from Hedra and third parties, with per-model schemas, limits, and pricing. Developers can estimate job costs before submitting, follow progress through polling, server-sent events, or signed webhooks, and stream lifecycle logs to their own endpoints. Hedra also offers World Model Runtime, model training, and GPU sandboxes as research previews, and Studio for creative work. Access is through a prepaid USD API wallet, with dedicated serving and training priced per engagement.
Create a v3 API key in the developer console with the scopes your integration needs, then add funds to the prepaid USD API wallet. Studio subscription credits are separate and do not fund v3 jobs.
Browse the model catalog and inspect a model's inputs, limits, and pricing with GET /v3/models/{model} before integrating, since model choice, dimensions, duration, and quality affect the charge.
For image, video, and audio jobs, send the same model-specific input to POST /v3/models/{model}/estimate. The estimate is a price, not a completion-time guarantee, and creates no job.
Submit to POST /v3/models/{model} and receive a 202 job envelope. Follow status with polling, the SSE stream, or signed webhooks, then retrieve outputs from GET /v3/jobs/{job_id}.
Send an image and a question to POST /v3/chat/completions, or use the Responses and Messages compatibility routes with the stock OpenAI and Anthropic SDKs, paid per token from the API wallet.
Use the Hedra MCP server to connect Claude, Cursor, Codex, or ChatGPT to Hedra tools for model discovery, cost estimates, supported media jobs, and library access.
Generating images, video, and audio programmatically through one API
Analyzing an image and answering questions about it with a vision model
Running open-weight LLMs and coding agents paid per token
Estimating and tracking inference spend from a prepaid wallet
Receiving job results in your own systems through signed webhooks
Connecting an AI assistant or terminal to Hedra tools via MCP
Serving and optimizing your own multimodal models on Hedra infrastructure
Creating avatar and lip-sync video from an image and audio input
Usage is billed from a prepaid API wallet in US dollars, per output and per token, with no subscription, seats, or minimum. Studio subscription credits are separate and do not fund v3 jobs.
Yes. For image, video, and audio jobs, send the same model-specific input to POST /v3/models/{model}/estimate. The estimate is a price, not a completion-time guarantee, and creates no job.
The catalog spans image, video, audio, and text models from Hedra and third parties, including Hedra Avatar, Hedra Character 3, FLUX, Kling, Veo, Seedance, ElevenLabs audio models, and open-weight LLMs such as GLM-5.3 and Kimi K3.
Submit to POST /v3/models/{model} and receive a 202 job envelope, then track progress with polling, the SSE stream at GET /v3/jobs/{job_id}/stream, or signed webhooks. Retrieve outputs from GET /v3/jobs/{job_id}.
Hedra exposes POST /v3/chat/completions plus Responses and Messages compatibility routes, so you can call open-weight LLMs with the stock OpenAI and Anthropic SDKs, paid per token from your API wallet.
Yes. The Hedra MCP server connects Claude, Cursor, Codex, and ChatGPT to Hedra tools for model discovery, cost estimates, supported media jobs, and library access.
World Model Runtime is Hedra's engine for multimodal workloads, purpose-built for diffusion, video, and world models. Hedra Training and GPU Sandboxes are currently research previews; contact the engineering team to discuss access.
Contact support at support@hedra.com, read the documentation at hedra.com/docs, or join the Hedra Discord community.
GPU compute is billed per GPU-second (GPU-hour equivalents shown). Training and GPU Sandboxes are research previews. Play monthly credits do not roll over and reset each billing cycle; credit-pack balances do not expire and are used after monthly credits. Play Ultra is priced from $100/month. Prices checked 2026-11-19.
GPU compute billed per GPU-second; equivalent to $8.25 per GPU-hour.
GPU compute billed per GPU-second; equivalent to $7.26 per GPU-hour.
GPU compute billed per GPU-second; equivalent to $2.49 per GPU-hour.
GPU compute billed per GPU-second; equivalent to $3.00 per GPU-hour.
GPU compute billed per GPU-second; equivalent to $0.89 per GPU-hour.
Play subscription with 2,000 credits per month.
Play subscription with 7,200 credits per month.
Play subscription starting from $100 per month with 18,000 credits per month; allowances up to 57,600 credits/month.
Play subscription with 14,400 credits per month.
Custom volume and pricing for enterprises.
COMMUNITY REVIEWS
No reviews yet.