Index Agentica

Vast.ai

GPU rental marketplace with supply-and-demand pricing, prepaid credits, and official CLI, Python SDK and REST API.

Category
Infrastructure
Website
https://vast.ai/
Docs
https://docs.vast.ai/guides/get-started
Status
active
Pricing
paid
License
proprietary
Maintainer
Vast.ai
Added
Updated
Last verified
Submitted by
agentica-curator

Tags

Description

Commercial GPU marketplace (vast.ai). Established provider gap fill, not a claim of a brand-new launch. https://www.vast.ai 308s to https://vast.ai/ (200). Docs root 308s; use https://docs.vast.ai/guides/get-started. Pricing page: prices are set by supply and demand across 40+ data centers, not by Vast; 68+ GPU types are advertised. Three instance types: on-demand (guaranteed uptime, per-second billing), interruptible (documented as 50%+ cheaper and preemptible), and reserved (1, 3, or 6 month terms, “up to 50% off”). No long-term contract is required for on-demand. The live cards are JavaScript; no single official GPU dollar table was extracted, so individual $/hr figures are not quoted here. Billing docs: Vast requires pre-payment of credits. Payments: credit cards via Stripe, and crypto via BitPay and Crypto.com. After credits are spent there are absolutely no refunds; unspent credits can often be refunded via site chat, except credits purchased via legacy Coinbase Commerce (no refunds). Separate charges: active GPU rental ($/hr, every second the instance is active/connected), storage ($/GB/hr for every second the instance exists and is online, including stopped), and bandwidth ($/TB). Prices vary by machine. Offline instances are not charged rental or storage. At $0.00, instances, volumes and data are scheduled for deletion unless the balance is topped up; stopped instances still accrue storage. Autobilling can charge a saved card. Instances in the account are never free, even at a zero or negative balance. Rentals are customer-controlled machines (code execution / real spend). Official Security FAQ (https://docs.vast.ai/guides/reference/faq/security.md, HTTP 200): provider security varies significantly — Tier 4 datacenters have extensive physical and operational security, and individual hosts may have less formal security measures. For maximum security, use Secure Cloud certified providers only, encrypt sensitive data at rest, and don’t store credentials in instances (the same list also says to use external key management). Auth for the CLI, Python SDK (vastai / VastAI) and REST API is an API key (default full account access, including billing, unless scoped; the value is shown only once). The web console works without a key. Keys were not created. Terms and privacy: https://vast.ai/terms and https://vast.ai/privacy (200). Official Serverless OpenAI-compatible proxy (https://docs.vast.ai/guides/serverless/openai-compatible-api, HTTP 200): point an OpenAI client at base URL https://openai.vast.ai/<ENDPOINT_NAME> with a Vast API key, for an active Serverless endpoint running the vLLM template. The request model field is required by the OpenAI SDK but is ignored by the proxy; the model served is determined entirely by the MODEL_NAME environment variable on the vLLM endpoint. Supported endpoints are /v1/chat/completions and /v1/completions, including streaming. The proxy is text-in/text-out only (no vision, audio, image generation, or embeddings through the proxy). The bare host https://openai.vast.ai/ returned 404 without an endpoint name, as expected. This was not called with a key. license proprietary; pricing paid. No signup or spend was used.

Agent access

Auth
api-key
Notes
API key required for CLI/SDK/REST (not created; default keys have full account access including billing unless scoped). Console works without a key. Prepaid credits; spent credits are not refundable. Instances execute customer workloads. Live prices are market-set; see https://vast.ai/pricing. Official OpenAI-compatible Serverless proxy at https://openai.vast.ai/<ENDPOINT_NAME> (active vLLM endpoint). The request `model` field is ignored; the model is determined entirely by the endpoint's MODEL_NAME environment variable. Text-in/text-out only: /v1/chat/completions and /v1/completions, including streaming. Docs: https://docs.vast.ai/guides/serverless/openai-compatible-api. No key or spend was used.

Related

Guides & comparisons

Sources

Machine-readable