The only one with a proof.

Great UX for comparing LLM price vs. speed vs. quality at a glance. Their calibration benchmarks are respected. But: everything is self-reported by labs, no replay, no proof.

They do run the evals themselves on HF hardware, the one trust anchor in the space. But scope is narrow (open-weight LLMs only), no tamper-evidence beyond "trust HF," and no dispute protocol.

Contamination-proof by construction (humans vote). But opaque methodology, subjective, can't be replayed, and impossible to certify any specific model's rank.

Every score on our leaderboard corresponds to a verified Ethereum transaction. Click the batch ID to see the proof on-chain. Fake a score, the proof fails, the listing is never published. No one else in AI evaluation has this.

LLMs aren't the whole stack. Memory layers, code agents, vector DBs, RAG, voice, sandboxes, MCP servers, we attest them all. A buyer of a modern AI product is choosing across all sixteen.

Six independent runners, each with posted ETH stake. If an upheld dispute shows they ran a benchmark wrong, the stake is slashed. Economic accountability that no one else has.

Every run publishes a one-liner that reproduces it on your hardware. We don't ask you to trust us, we hand you the commands to not trust us.