Platform
The control plane for open-weight AI.
One endpoint for serverless models, dedicated GPUs, fine-tunes, and governance — so product teams ship without each building their own provider stack.
Live models
11
Providers
10
Gateway
opendoor-gcp.web.app
What you can build
Build anything on OpenDoor.
Serverless open models
Call Llama, Qwen, DeepSeek, and more with published $ / 1M rates — no Request GPU step. Wholesale warm path when Together is configured.
See ratesDedicated deployments
Run on this Mac (Ollama / Metal) or GCP Cloud Run GPU. Custom weights, precision, autoscaling, scale-to-zero, and reserved capacity.
Open deploymentsMulti-LoRA & routers
Load adapters on vLLM endpoints and split traffic with A/B routers (router:slug) across healthy deployments.
Open routersFine-tunes at base price
Upload datasets, run SFT / DPO / ORPO jobs, and serve ft: models billed at the base model list price.
Open trainingGoverned access
Scoped API keys, spend caps, service_tier priority, and org policies enforced before every provider call.
Security overviewObservability & billing
Tokens, latency, cached rates, GPU-seconds, and usage explorer in one dashboard — Stripe top-ups included.
Open usageWhat you get day one
Fireworks-style commercial surface plus OpenDoor governance. Worldwide residency — process your data where you want, and we can support it.
OpenAI-compatible API
Chat, completions, embeddings, rerank, vision, structured outputs, and batches on /v1.
GPU SKUs
Standard, Fast, and Ultra GPUs available — on-demand, with per-second metering for dedicated fleets.
Trust center
Model approvals, policies, violations, sector packs, and immutable audit logs.
Create a free workspace
Top up $20, get $5 of open-weight credit, and call a live model from the playground.