BlockRun
Per call. Per token. No subscription.

Agents can think.
We let them spend.

Chat, image, video, voice — plus the data, tools, and compute an agent needs to finish the job. One endpoint, and every call pays for itself in USDC.

Live / on-chain, Base
18,300,000
transactions settled
ModelAmountAge
Listening for on-chain payments…
86 models · 100 data & tool APIsVerify on Basescan ↗Live feed →
BlockRun MCP · in the terminal

Watch an agent
pay its way.

A Claude Code session with BlockRun MCP. The agent asks, the call settles in USDC, the receipt's on-chain.

claude code · mcp
get a second opinion on this race condition from GPT-5.6
blockrun_chat(model: "openai/gpt-5.6-sol", prompt: "review queue/worker.ts — workers occasionally double-claim jobs…")
402 → signed USDC auth → verified
"Two workers can read status='pending' before either UPDATE lands. Move the claim check inside the row lock — SELECT … FOR UPDATE SKIP LOCKED."
$0.0042 · settled on Base · 0x8f2a…c41d
AI model gateway

86 models.
One endpoint.

Chat, image, video, music — one OpenAI-compatible API. You pay the provider's rate plus 5%, and we publish the provider's rate.

FreeTry without paying — no card, no signup
modelprovin / out · 1M
Claude Fable 5
anthropic$10.00 / $50.00
GPT-5.6 Sol
openai$5.00 / $30.00
Claude Opus 5
anthropic$5.00 / $25.00
Claude Sonnet 5
anthropic$3.00 / $15.00
Gemini 3.1 Pro
google$2.00 / $12.00
Gemini 3.5 Flash
google$0.50 / $3.00
Enterprise

AI infrastructure
for teams.

Spend caps that hold, usage you can attribute to a team, and an audit trail that reconciles against the chain.

Budget controls
Per-key spend caps, hard limits, alerts
Usage analytics
Per-team, per-model, per-call attribution
Team management
SSO, roles, audit log, key rotation
Dedicated support
Slack channel, SLA, named CSM
We don't share your data

No training, no retention beyond the request. Privacy policy →

No accounts, no KYC

Wallet in, prompt out. Pseudonymous by default.

Open-source SDKs, MIT

Audit the wire format on npm and PyPI.

Ship today

Point your client at BlockRun.
The first call already works.