NexBench tests blockchain AI agents across 214 tasks
NexBench evaluates whether an can complete real on-chain work. It contains 214 tasks across eight and runs every task five times to check .
The tasks cover executing transactions, choosing swap routes, bridging funds between blockchains, managing DeFi positions, researching tokens, detecting a drainer, rebuilding portfolio records, and carrying out treasury . Tests run in forked-mainnet environments, which reproduce the chain conditions without relying on live network state.
Key points
- The benchmark contains 214 tasks in eight .
- Every task is run five times under repeatable conditions.
- It tests transactions, swaps, cross-chain transfers, DeFi management, and threat detection.
- It also covers portfolio reconstruction and treasury .
- The tests use forked-mainnet environments.