SWE-bench Verified
~54%Token Reduced
The Inference CDN for AI AgentsCompute once, serve everywhere.
Built on Global Infrastructure. Designed for Persistent Intelligence.
Hit Rate
~81%
Context Compression Rate
~14.84x
Error Rate
As low as<0.5%
Agents pay FULL PRICE for repeated content
24:1
Token imbalance
Coding agents send up to 24x more tokens in than out
Repeated context
Same input, full price
The same files and questions hit the model again and again — at full price
Provider-only
Vendor cache limits
Provider caches are single-vendor, exact-prefix, and short-lived
The web solved this with CDNs.
Inference hasn't — until now.
One
endpointThreesavings mechanisms
Cache
Semantic cache with a user-set error bound δ. Repeated requests never reach the origin.
Compression
Action-preserving context compression for coding agents. Same next action, fewer tokens.
Settle
Every hit and every saved token is metered — you pay 20% of list price on what you save.
Coding Agents
For individual, Infergrid is plug-and-play.
Claude Code
ANTHROPIC_BASE_URL="https://t.infergrid.network/anthropic"
ANTHROPIC_AUTH_TOKEN="sm_live_key"
Savings you can verify on-chain
We take no margin on model prices.
Our revenue comes only from savings we can prove.
LIVE
Stablecoin usage-based billing
IN DEVELOPMENT
Signed savings receipts anchored on-chain
DESIGN STAGE
Distributed cache-node network with x402 settlement
No token. Incentives launch only where metered demand already exists.
Read Whitepaper §7–8.Start saving in 30 seconds
One Line. No migration. Cancel anytime.