The Winged Victory of Samothrace landing on the prow

Backed by HF0

Agents spend money. We measure.

9 of 12 frontier models got scammed with real money. See the leaderboard →

Agent Money Benchmark · adversarial trackOct–Nov 2026
#ModelTask successScam rate
1Gemini 4 Argon98.8%0.8%
2GPT-6 Astra98.8%0.0%
3Claude Fable 5.298.8%0.8%
4GPT-6.1 Sol97.5%0.0%
5Claude Opus 5.597.5%0.0%

Agents buy data with a client’s money; some sellers cheat. Graded from the payments, not by a model. Method ↗

Winged Victory
of Samothrace ·
Louvre · CC0
53M payments settled on the ledger behind every score

One goal behind all three

Proof that agents handle money right.

We settle what agents pay for, so we see every payment they get wrong and what each dollar bought. All three products are built on that.

  1. Choosing

    Which setup to trust with money: paid right, kept to its limits, and what each completed task cost.

    01 Verified Eval
  2. Training

    Agents that learn to make better decisions with money.

    02 RL environments
  3. Routing

    Developer infrastructure for pay-per-call AI models, tools, and data. Pay for the outcome.

    03 Developer API
Settlement

The common foundation behind all three: the same settled payments, read three ways.

Three ways to work with us

The prow of the ship beneath the Winged Victory

Financial institutions

The model and setup that make your product better.

We run your own production cases across models and harnesses and grade each answer with code. Where the agent moves money, we check it paid the right party the right amount, within its limits, and said so truthfully. You see which combination holds up, and what it costs.

Verified Eval →
The wing and shoulders of the Winged Victory, the head never recovered

AI labs

RL environments for agents that handle money.

Built to order: agents that pay for real, scored by the settlement ledger instead of a judge. Every task synthesized by us; no customer data.

For AI labs →
The foot of the Winged Victory touching the prow

Developers

Pay per call.
Two ways in.

One endpoint, 124 models, each call priced before it runs. Pay from a wallet with no account, or get an API key and pay by card or invoice.

For developers →

Build with BlockRun

One infrastructure layer for AI teams and developers.

For enterprise

Prove your agents handle money right, and what they cost.

Talk to our team

For developers

Access models and start building.

Explore the docs