Next.js Evals

Next.js Evals is a system designed to test and measure how well different AI coding agents perform when working with Next.js, a popular web development framework. These evaluations focus on tasks like creating new Next.js code and updating existing code to newer versions.

Benefits

Next.js Evals helps developers understand which AI tools are most effective for Next.js projects. By providing clear performance data, it allows users to choose the best AI agent for their coding needs, potentially saving time and improving code quality.

Use Cases

This system is useful for developers who want to integrate AI assistance into their Next.js workflows. It can help in deciding which AI coding assistant to use for tasks such as generating boilerplate code, refactoring components, or migrating projects to the latest Next.js features. The results show how different AI models stack up against each other in these specific areas.

Vibes

Based on the evaluation results from February 10, 2026, AI agents showed varying degrees of success. For instance, the GPT 5.3 Codex agent achieved a 90% success rate in Next.js evaluations, while other agents like Claude Opus 4.6 and Gemini 3.0 Pro Preview had success rates of 80% and 75% respectively. Some agents performed at lower rates, with success rates as low as 35% for GPT 5.2 Codex and 40% for Minimax M2.1 and Devstral 2.

Additional Information

The evaluations were performed on February 10, 2026, using a set of 20 evaluations for each AI agent tested. The agents included in the report are Codex, Claude Code, OpenCode, Cursor, and Gemini CLI, powered by different AI models like GPT, Claude, and Gemini.