Vast.ai API

Vast.ai is a peer-to-peer GPU marketplace where buyers rent idle consumer and datacenter GPUs from host machines at market-rate prices, accessible via a REST API for automated instance search, bidding, and management.

Evaluated Mar 06, 2026 (0d ago) vcurrent
Homepage ↗ Other gpu marketplace cheap-gpu spot-gpu ai-training inference vast-ai peer-to-peer
⚙ Agent Friendliness
51
/ 100
Can an agent use this?
🔒 Security
69
/ 100
Is it safe for agents?
⚡ Reliability
62
/ 100
Does it work consistently?

Score Breakdown

⚙ Agent Friendliness

MCP Quality
--
Documentation
70
Error Messages
62
Auth Simplicity
83
Rate Limits
55

🔒 Security

TLS Enforcement
100
Auth Strength
68
Scope Granularity
45
Dep. Hygiene
65
Secret Handling
68

Instances run on third-party consumer hardware with no formal vetting of hosts. Data isolation is only as strong as Docker container boundaries. No compliance certifications. API key is a single credential with full account access.

⚡ Reliability

Uptime/SLA
55
Version Stability
68
Breaking Changes
65
Error Recovery
60
AF Security Reliability

Best When

You need the absolute lowest GPU rental cost and can tolerate variability in hardware quality and reliability that comes with a peer-hosted marketplace.

Avoid When

Your workload requires consistent hardware configurations, certified data isolation, or any form of uptime guarantee.

Use Cases

  • Search GPU offers by price-per-TFLOP and automatically rent the cheapest available instance matching VRAM and CUDA requirements
  • Automate training job dispatch by creating instances, monitoring status, and destroying them on completion
  • Run cost-optimized batch inference by bidding on spot instances and submitting jobs programmatically
  • Retrieve real-time offer listings to build dynamic cost-comparison dashboards for GPU pricing
  • Manage multiple concurrent rented instances — start, stop, and label them — from within orchestration scripts

Not For

  • Production inference requiring guaranteed uptime or SLA-backed availability
  • Compliance-sensitive workloads where hardware provenance and data isolation must be certified
  • Teams needing predictable, reservation-based GPU access without auction/bidding mechanics

Interface

REST API
Yes
GraphQL
No
gRPC
No
MCP Server
No
SDK
Yes
Webhooks
No

Authentication

Methods: api_key
OAuth: No Scopes: No

API key passed as query parameter (?api_key=) or in the Authorization header. Single key per account with no scope differentiation.

Pricing

Model: pay-as-you-go
Free tier: No
Requires CC: Yes

Prepaid credit system. Spot (interruptible) instances are cheaper than on-demand. Prices fluctuate with marketplace supply and demand.

Agent Metadata

Pagination
offset
Idempotent
Partial
Retry Guidance
Not documented

Known Gotchas

  • Host machines are consumer hardware — instances can disappear without warning if the host goes offline, requiring agents to detect and re-provision
  • Instance 'created' status does not mean SSH is ready; agents must poll until the instance is actually reachable, which can take 2-10 minutes
  • Offer search results change rapidly; an offer selected in one API call may be gone by the time the create call fires
  • No native job queuing — agents must implement their own retry logic when desired GPU specs are unavailable
  • Data on rented instances is not preserved after destruction; agents must explicitly copy results out before terminating

Alternatives

Full Evaluation Report

Detailed scoring breakdown, competitive positioning, security analysis, and improvement recommendations for Vast.ai API.

$99

Scores are editorial opinions as of 2026-03-06.

5178
Packages Evaluated
26151
Need Evaluation
173
Need Re-evaluation
Community Powered