🧪Benchmark Tests Five JavaScript Frameworks by Having Agents Build a Pokémon App

Your framework pick for agents might be vibes

Everyone says their framework is better for agents. Nobody has been measuring it. A new benchmark finally did, and the verdict is more useful than the hype. Here's the setup: agents were told to build a small but non-trivial Pokémon app on PokéAPI using a specific JavaScript framework or meta-framework. Each output was checked against a spec conformance suite so agents couldn't cheat their way to a passing grade. Results were scored on development time, tokens spent, lines of code, and Lighthouse performance and accessibility. The cost spread is real. Runs landed between $2.67 and $4.97, with a median of $3.14. SvelteKit, Gracile+Lit, and Angular cost more to build than the rest. If you're picking a stack for agent-driven work, that's the number to plug into your budget, not a tweet from a framework maintainer. The catch: this is one simple app, so it represents a single holotype. Not every framework had an MCP server or Claude Code plugin, and agents used them unevenly, so tooling skew could move the numbers.