Gateway | VLM Run
ProductPlatform<br>GatewayOne OpenAI-compatible API for every visual model.<br>Orion<br>Chat PlaygroundTry Orion in our hosted chat interface.WhitepaperThe technical overview of Orion, our flagship visual agent.<br>Open Source<br>mmFast, multi-modal context for agents.vlmbenchBenchmark any VLM on your hardware, in one command.vlmrun-hubStructured schemas for production VLM use cases.
SolutionsBy Modality<br>DocumentsTurn hours of manual document work into seconds of schema-validated JSON.ImagesComing soonCaption, detect, segment, and generate. Every image operation through one API.VideosComing soonSummarize, transcribe, and search long-form video without watching it.<br>By Industry<br>Physical AIAgentic data labeling for demonstrations, reconstruction, and grounded training data.HealthcareTurn messy faxes, scans, and clinical paperwork into clean, validated data.ConstructionTurn blueprints, schedules, and specs into structured data.
PricingDocsBlogAboutBook a DemoDashboard
One unified API, Any Visual Model.<br>Document OCR, captioning, and multi-modal chat: every visual capability behind one MCP server.<br>Read the DocsGet an API Key
Pydantic AIMastraLangChain
from pydantic_ai import Agent<br>from pydantic_ai.capabilities import MCP
mcp = MCP(<br>"https://gateway.vlm.run/mcp",<br>headers={"Authorization": "Bearer "},<br>agent = Agent("anthropic:claude-sonnet-5", capabilities=[mcp])
print(agent.run_sync("Extract the invoice fields from invoice.pdf").output)
Supported providers
Explore models
7 visual models. One endpoint.
AllChatCaptionOCRMarkdownDetectionSegmentationPoseSegmentation
AllChatCaptionOCRMarkdownDetectionSegmentationPose<br>Search modelsAll Providers
ModelCapabilitiesContextInput / 1MOutput / 1Mdots.mocr
rednote-hilab/dots.mocr
markdown<br>32K$0.15$0.30Florence-2
microsoft/florence-2-base-ft
ocrdetectioncaption<br>—$0.01$0.01PP-OCRv6
paddleocr/pp-ocrv6
ocr<br>—$0.01$0.20DeepSeek-OCR-2
deepseek-ai/DeepSeek-OCR-2
markdownocr<br>32K$0.06$0.12GLM-OCR
zai-org/GLM-OCR
markdown<br>8K$0.10$0.20PaddleOCR-VL
paddlepaddle/paddleocr-vl
markdownocr<br>16K$0.05$0.10Qwen3.5 0.8B
qwen/qwen3.5-0.8b
chatcaptiondetection<br>262K$0.08$0.15LightOnOCR<br>Coming soon<br>lightonai/lightonocr-1b-1025
markdown<br>-——Unlimited-OCR<br>Coming soon<br>baidu/unlimited-ocr
markdownocr<br>———Qwen3.8 27B<br>Coming soon<br>qwen/qwen3.8-27b
chatcaptiondetection<br>———Gemma 4 26B-A4B<br>Coming soon<br>google/gemma-4-26b-a4b
chatcaptiondetection<br>———SAM 3<br>Coming soon<br>facebook/sam3
segmentation<br>———RF-DETR Large<br>Coming soon<br>roboflow/rf-detr-large
detection<br>———ViTPose+<br>Coming soon<br>usyd-community/vitpose-plus-small
pose<br>———<br>14 of 14 models
Pricing calculator<br>Dirt-cheap Document OCR.<br>Pick a document type, page volume, and OCR model. See cost savings versus closed vision APIs.
Document typeLegal: ~1000 tokens / pageHealthcare: ~700 tokens / pageConstruction: ~600 tokens / pageSlides: ~400 tokens / pageForms: ~250 tokens / page
OCR modelAll models (price band)VLM Run Gateway (dots.mocr)VLM Run Gateway (GLM-OCR)VLM Run Gateway (DeepSeek-OCR-2)VLM Run Gateway (PP-OCRv6)
100K
Page volume / month
Gateway OCR tokens (in / out)<br>120M - 250M / 100M<br>Frontier billed tokens (in / out)<br>250M / 200M<br>2.5K image tokens per page; output includes reasoning at 2× OCR text.
Total cost savingsfrom frontier models<br>$7.2K - $7.2K
You savefrom frontier models<br>99 - 100%/mo
Documents / month<br>100Kpages
Cost comparison<br>VLM Run Gateway<br>Cheapest - most expensive
$19 - $68
OCR APIs<br>Textract, Azure Doc AI
$150 - $1K
Document AI APIs<br>Reducto, LlamaParse, Extend
$1K - $6K
Frontier VLMs<br>Gemini, Claude, GPT
$938 - $7.3K
$0$1.8K$3.6K$5.4K$7.3K
Estimates vs typical OCR, Document AI, and frontier VLM pricing.Source: llm-prices.com
A vision-only gateway, built for builders.<br>LLM routers and gateways route to 100s of LLMs, yet only a handful of VLMs, and often no OCR or classical CV models. Visual AI deserves its own stack.
01
Multimodal, Multitask<br>One catalog spanning multi-modal inputs and multi-task outputs: OCR, detection, segmentation, pose, keypoints, and more.
02
Chat Completions Native<br>OCR, captioning, and multimodal chat run through the same OpenAI-compatible chat completions API you already use.
03
Orchestration Built-In<br>Send a 500-page PDF or a 2-hour video in a single call. The gateway chunks, batches, and reassembles for you. No pipelines to build.
04
Structured Outputs, Out of the Box<br>JSON-schema enforcement on every call. CV wrappers emit fixed schemas; VLMs honor response_format.
05
Open Models, No Lock-In<br>Every model is open-weight, served through the OpenAI-compatible API you already use. Swap models or providers freely, your client code never changes.
06Coming soon
Agent-native vision, via MCP<br>Give any MCP client instant access to the full visual model catalog. Agents see, read, and reason over images out of the box.
Built for production visual AI.<br>Optimized for production workloads and cost-efficiency. Every model deployed gets its own performance tune-up.
Models...