BlockRun
For developers

Pay per call. No account.

One endpoint, 115 models. ClawRouter picks the cheapest one that completes the task.

$ pip install blockrun-llm
$ export BLOCKRUN_WALLET_KEY=0x… # signs locally, never sent
$ python -c "from blockrun_llm import LLMClient; print(LLMClient().chat('openai/gpt-5.2', 'Hello!'))"
→ 402  priced before it runs
→ 200  settled · Hello! How can I help?
The foot of the Winged Victory touching the prowlanding · 5.6°

What runs it

01 · ClawRouter

The cheapest model that completes the task.

MIT. Classifies each request locally in under a millisecond. Published profiles save up to 84 %.

02 · BlockRun MCP

Claude Code, Cursor, OpenClaw, ChatGPT.

Tools for the harnesses people use. The call pays for itself.

03 · SDKs

Python and TypeScript.

blockrun-llm on PyPI, @blockrun/llm on npm. MIT.

04 · Models

Every model, with its per-call price.

Chat, reasoning, coding and vision; image, video, music and speech. Free open-weight models to try without paying.