What do you want to build?
Varity AI Gateway is a private AI chat that builds, deploys and manages your apps. 183 models. Your prompts and replies are never stored.
Playground
Ask anything, then ship what you build
The Playground is a private AI chat that also builds your app, deploys it to Varity Cloud and manages it. Nothing changes until you confirm.
For the work you can’t paste into other chatbots
PlaygroundNew chatExampleMASTER SUPPLY AGREEMENT
Page 3
4.2 Renewal
This Agreement renews automatically for successive twenty-four (24) month terms unless either party gives written notice of non-renewal at least ninety (90) days before the end of the then-current term.
6.3 Pricing
Supplier may revise its prices on thirty (30) days’ written notice to Customer.
8.1 Confidentiality
Each party shall hold the other party’s Confidential Information in confidence and use it only to perform this Agreement.
9.1 Limitation of liability
Supplier’s total aggregate liability shall not exceed the fees paid by Customer in the one (1) month preceding the claim.
Your prompts aren’t our product
Varity keeps a usage record of each request for 30 days. Never what you asked, and never what the model answered.
Private86 models
Your prompts and replies aren’t stored or used for training.
Including Qwen 3.8 27B, GLM 5.3 Flash and Grok 4.7
Anonymized97 models
Sent without your identity.
Including GPT-6 Luna, Claude Opus 5.5 and Qwen 3.8 Flash
Require Private for your account, one API key or a single request. Varity never falls back to a weaker class.
Request details
req_7f3a91c2d8e4c21e
Model
Qwen 3 Coder 480B Turbo
Canonical ID: qwen-3-coder-480b-turbo
API key
support-bot
Timing
1,840 ms
TTFT 380 ms
Tokens
1,240
912 input · 328 output
Cost
Provider cost
What the model service charges, nothing added.
Privacy
Private
Retention
Metadata 30 days
Prompts and responses not stored by Varity.
Prompt
Never stored
Response
Never stored
183 models, provider prices
From 43 makers. Chat, image, speech, transcription and embedding models on one AI Gateway key. You pay what the model service charges, with nothing added per request. A 5% fee applies when you add credits, from $20.
Aion 3.5Anonymized$3.75 in · $7.50 out / 1M
Claude Opus 5.5Anonymized$4.80 in · $24.00 out / 1M
DeepSeek V4.1 FlashPrivate$0.375 in · $1.50 out / 1M
Mercury 2.5Anonymized$0.05 in · $0.1875 out / 1M
GLM 5.3 FlashPrivate$0.15 in · $0.50 out / 1MInklingTranscriptionPrivate$1.25 in · $5.06 out / 1MKrea 2 TurboImagePrivate$0.04 per image
MiniMax M3 PreviewPrivate$0.30 in · $1.20 out / 1MGradium TTSSpeechAnonymized$47.50 per million charactersIdeogram V4ImageAnonymized$0.06 per imageChatterbox HD (Resemble AI)SpeechPrivate$50.00 per million charactersInworld TTS-1.5 MaxSpeechAnonymized$12.50 per million charactersOrpheus TTSSpeechPrivate$62.50 per million charactersGemma 4 UncensoredPrivate$0.1625 in · $0.50 out / 1MFireRedImagePrivate$0.04 per imageBackground RemoverImageAnonymized$0.03 per imageRecraft V4ImageAnonymized$0.05 per imageChromaImagePrivate$0.01 per image
Mistral Small 3.2 24B InstructPrivate$0.0938 in · $0.25 out / 1M
Hermes 3 Llama 3.1 405bPrivate$1.10 in · $3.00 out / 1MSD35ImagePrivate$0.01 per imageAion 3.5 MiniAnonymized$0.875 in · $1.75 out / 1MClaude Fable 5.1Anonymized$12.00 in · $60.00 out / 1MDeepSeek V4 Pro 0813Private$1.65 in · $4.95 out / 1M
GPT-6 LunaAnonymized$0.125 in · $0.625 out / 1M
Grok 4.7Private$2.27 in · $6.80 out / 1M
Qwen 3.8 FlashAnonymized$0.14 in · $0.49 out / 1M
Gemini 3.8 FlashTranscriptionAnonymized$0.9375 in · $4.69 out / 1M
Kimi K3 FastPrivate$4.50 in · $22.50 out / 1M
Seedream V5 ProImageAnonymized$0.06 per imageLuma Uni-1ImageAnonymized$0.05 per image
MiMo-V2.5TranscriptionPrivate$0.40 in · $2.00 out / 1M
NVIDIA Nemotron 3 UltraPrivate$0.625 in · $3.13 out / 1MBGE-EN-ICLEmbeddingsPrivate$0.0125 in / 1MElevenLabs Scribe V2TranscriptionAnonymized$0.000167 per audio secondMultilingual E5 Large InstructEmbeddingsPrivate$0.0125 in / 1MWizper (Whisper v3)TranscriptionPrivate$0.0001 per audio secondUncensored 1.2Private$0.20 in · $0.90 out / 1MHunyuan Image 3.0ImagePrivate$0.09 per imageRole Play UncensoredPrivate$0.50 in · $2.00 out / 1MGLM 4.7 Flash HereticPrivate$0.07 in · $0.40 out / 1MImagineArt 1.5 ProImageAnonymized$0.06 per imageFlux 2 MaxImageAnonymized$0.09 per image
Llama 3.3 70BPrivate$0.70 in · $2.80 out / 1MKokoro Text to SpeechSpeechPrivate$3.50 per million charactersGPT-6 SolAnonymized$2.50 in · $12.50 out / 1MGrok Imagine 2.0ImagePrivate$0.07 per imageQwen 3.8 27BPrivate$0.45 in · $3.20 out / 1MAPI
Keep your OpenAI SDK
It’s an OpenAI-compatible API: change the base URL and key. For a team, give each person or app its own key that can require Private models, allow only the models you choose, and expire on its own.
Create an inference key in the Developer Portalimport osfrom openai import OpenAI client = OpenAI( api_key=os.environ["OPENAI_API_KEY"], base_url="https://ai.varity.app/v1", api_key=os.environ["VARITY_API_KEY"],) response = client.chat.completions.create( model="<a model from the catalog>", messages=[{"role": "user", "content": "Hello"}], extra_body={"varity_extensions": {"privacy_floor": "private"}},)Questions, answered
Does Varity store my prompts?
No. Varity stores no prompts, responses, tool arguments, uploads or Playground conversations. It keeps a usage record of each request, such as time, model, tokens and cost, for 30 days.
Which models are Private?
86 of the 183 live models are Private, including Qwen 3.8 27B, GLM 5.3 Flash and Grok 4.7. Anonymized models, including GPT-6 Luna, Claude Opus 5.5 and Qwen 3.8 Flash, are sent without your identity. Each model shows its class before you send.
What is the difference between Private and Anonymized?
Private: your prompts and replies aren’t stored or used for training. Anonymized: sent without your identity. Choose Private for your account, an API key or a single request, and Varity never falls back to a weaker class.
What can the Playground do?
Chat with a model from the live catalog, attach files, search the web and compare models side by side. With Varity tools on, it can also read your repository, deploy your app and manage your deployments. Every change waits for your confirmation.
Can I control what my team uses?
Yes. Each API key can require Private models, be limited to the models you approve, and be revoked at any time. You can also cap monthly spend.
How is it priced?
Each request costs what the model service charges, with nothing added. A 5% fee applies when you add credits, from $20 up to $1,000 at a time. There is no signup credit, and bring-your-own-key is not available.
Can I use my existing OpenAI SDK?
Yes. Set the base URL to https://ai.varity.app/v1, use your Varity API key and a model ID from the catalog. Streaming and tool calling work on eligible models.
Can I run my own model?
AI Gateway serves the models in its catalog. To run your own model or a training job, use Varity GPU compute.
Start a private chat
Open the Playground and send your first message