Vercel's Eve Agent Framework Reaches 1.0, With Mixed Coding Benchmark Results
Vercel released Eve 1.0 as an Apache-2.0 framework for building and operating agents. Eve supplies the runtime around a model, including tools, workflow state, and delegation. Developers can choose a model and keep agent behavior and permissions in their own codebase.
What the benchmark says
In the coding-extension comparison quoted in the source article, with the same tasks and model, Eve achieved 69% correct, against 81% for OpenCode and 78% for fx. That comparison is useful if you are choosing a coding harness. It does not cover every kind of agent or deployment.
Review the defaults before deployment
The documentation cited by the Chinese article describes a default configuration that does not require human approval for tool calls and does not deny all outbound sandbox traffic. Use only the tools and network destinations required for the task, and add approval for changes with external consequences.
Where Eve fits
Eve is worth evaluating when your team wants to own the runtime, tools, and approval logic. Benchmark against a simpler workflow first, with the same model, tasks, and review rules. Compare latency, per-task cost, inspectability, and recovery as well as accuracy.