OnchainPlanBench Seed: Web3 AI-Agent Action Plan Evaluations

CodePit released OnchainPlanBench Seed, an early dataset for evaluating small open-weight models. The dataset tests a model's ability to critique, repair, reject, or approve Web3 AI-agent action plans before wallet execution. It was last updated on June 2, 2026.

Use Cases

• Benchmarking model performance on plan critique tasks based on provided user intent and wallet context.
• Training models to approve or reject action plans based on provided risk and privacy policies.
• Developing tools for plan repair based on the available tools and policy constraints described in each row.

Strengths

• Public seed release for the first official CodePit model track, CodePit PlanGuard.
• Designed to test multiple safety functions: critique, repair, rejection, and approval.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

CodePit

Freshness

Last updated 2026-06-02 12:16:34; freshness should be verified.