01
What VLM Run Gateway is
VLM Run Gateway is a unified API for visual AI workloads including OCR, multimodal chat, document extraction, detection, segmentation and other vision tasks. VLM Run positions it as a vision-specific alternative to general LLM routers.
The public catalog is dynamic. Model names, modalities and prices can change, so production integrations should validate the current catalog instead of treating launch-day examples as permanent.
02
Why VLM Run built a vision-specific gateway
VLM Run says its internal testing exposed vision-specific serving problems around quantization, small-text OCR quality, video inputs, FPS controls and document orchestration. Those observations are vendor-reported product rationale rather than independent industry benchmarks.
The Gateway attempts to absorb document rasterization, page batching, retries and reassembly so callers can focus on choosing a model and validating output instead of rebuilding the serving pipeline for every provider.
03
OpenAI-compatible API and CLI
VLM Run documents an OpenAI-compatible base URL at https://gateway.vlm.run/v1/openai/. Its examples use the OpenAI SDK, a VLM Run API key when authenticated access is used, and a model ID selected from the current catalog.
The launch docs also show a CLI workflow using uvx vlmrun gw models to inspect the catalog and uvx vlmrun gw chat for document, image and video tests.
04
Documents, OCR and Markdown extraction
Document workflows can send a PDF or document input to a compatible OCR or multimodal model and request structured text or Markdown. VLM Run says the Gateway can chunk, batch and reassemble large document jobs internally.
That can simplify invoice extraction, contract ingestion and RAG pipelines, but output quality still depends on the selected model and the source material. Teams should test tables, small text, scans and multilingual pages on their own corpus.
05
Video support and MCP for agents
VLM Run also targets image and video understanding, but capability varies by model. A model that accepts images should not automatically be assumed to support native video or the same sampling controls.
The Gateway is exposed through MCP so compatible agents can access visual models without a separate integration for each model family. MCP support is part of the product positioning, while exact setup details should be checked in the current docs.
06
Current alpha access and pricing
VLM Run's September 2026 Gateway launch page says anonymous access is available with IP-based rate limits and that the Gateway is free during alpha. The same page says published prices are already live as cost signals for production.
Separately, VLM Run's current pricing page lists a Starter plan at $0 per month with pay-as-you-go usage and a $10 signup balance, plus paid Pro and Enterprise tiers. These are distinct states: alpha access should not be described as permanently free, and account billing should be checked before production use.
07
Enterprise deployment and limitations
VLM Run advertises private VPC or on-premises deployment, SSO, RBAC, audit-log export and regulated-data options. Organizations should verify the exact contract, retention policy and compliance scope for their own deployment rather than relying on marketing labels alone.
- model availability and pricing can change during alpha
- not every model supports every modality
- vendor quality, latency and savings claims require independent validation
- underlying open-weight models can carry different licenses
08
Who should consider VLM Run Gateway
The Gateway is most relevant to teams building document ingestion, OCR-heavy automation, multimodal agents, visual search, asset tagging or video-understanding systems that need to compare multiple visual models behind one interface.
Sources
Primary and supporting sources
Facts were rechecked against the linked sources immediately before publication. Pricing, product availability and rollout status can change.