Frontier models.
Zero friction.
Hold $MODEL, mint a key, point your client at one endpoint. Your live balance sets the tier — no invoices, no seats, no snapshots.
curl https://api.model-network.xyz/v1/chat/completions \
-H "Authorization: Bearer $API_KEY" \
-d '{"model":"api/atlas-large",
"messages":[{"role":"user","content":"Hello"}],
"stream": true}'How it works
Your balance becomes your plan.
Four steps from wallet to first token. Nothing custodial, nothing manual.
Connect a wallet
Non-custodial sign-in. We read your balance, never hold your keys or your tokens.
Your tier resolves live
The balance is read on-chain at request time and mapped to a quota and rate limit.
Mint an API key
Scoped keys per project. Rotate or revoke at any time from the developer console.
Call one endpoint
OpenAI-compatible chat completions with streaming. Every call is metered and receipted.
Models
The frontier, at your fingertips.
View all models →Product
Three surfaces, one access layer.
Keys & tiers
Hold at least 10,000 $MODEL and your key works immediately. Drop below the minimum and usage pauses until it qualifies again.
Wallet & tiers →Guarded agents
Research, coding, data and browser agents run behind a policy engine — spend caps, runtime limits, allow-listed tools and domains.
See agents →Real accounting
Every call is attributed to a wallet, account and key: model, tokens, latency, cost and status — visible the moment it lands.
View activity →The numbers
Access scales with what you hold.
Tiers read your live on-chain balance. Holdings above 30M $MODEL are recognized at the cap.
- ≥ 100K $MODEL held
- 10 requests / min
- All models & agents
- ≥ 1M $MODEL held
- 30 requests / min
- All models & agents
- ≥ 5M $MODEL held
- 90 requests / min
- All models & agents
- ≥ 20M $MODEL held
- 300 requests / min
- All models & agents
GPU Compute
H100, H200, A100 and RTX 5090 capacity — rent raw GPUs or list your own and earn. The rails are built; access opens after the API platform.
Ship on the Model Network
Create an account, link a wallet, mint a key — and point any OpenAI-compatible client at our endpoint.