Frameworks and backends
Official framework evals, plus one ecosystem eval. Read each board on its own. The baseline is the column without skills, guidelines, or AGENTS.md.
Library 4 · Skill 1 · Prompt 2 · Edited 2026-10-10
All
Next.jsLibraryThe full-stack React framework. The official agent evals use pass@4 for Next.js tasks.Fit：A web app that wants routing, server rendering, and the Vercel deploy path.MIT · nextjs.orgNext.js AGENTS.mdPromptThe bundled Next.js docs in the eval. Each model's change is listed on its own, including zeros.Fit：The repo uses Next.js and the agent should read official docs instead of writing the App Router from memory.nextjs.org/evalsReactLibraryThe UI library. ReactBench is run by Million. It is an ecosystem eval, not an official React or Meta board.Fit：Component UI. When you also use Next.js, the framework eval is on the Next.js page.MIT · react.devSupabaseLibraryPostgres, auth, and storage. The official eval asks whether a model can finish tasks on this backend.Fit：A full-stack app that needs a database, login, and files without running Postgres yourself.MIT · supabase.comSupabase skillsSkillThe supabase and supabase-postgres-best-practices skills loaded in the eval. The change is listed per model, including zeros and drops.Fit：An agent will write Supabase queries, RLS, or migrations, and you want the official skills loaded.supabase.com/evalsConvexLibraryA backend with queries and mutations. The official default rank is no guidelines and no web access.Fit：A TypeScript app that wants data, functions, and client types together.Apache-2.0 · convex.devConvex AI rulesPromptThe agent rules in the Convex docs. With Guidelines scores are unverified, so this page does not list a change.Fit：The project uses Convex and the agent should follow the official rules for queries and mutations.docs.convex.dev/ai#convex-ai-rules
How to choose
Read the baseline first Rank uses the column without skills, guidelines, or AGENTS.md.
Then read the change Skill and rules pages list each model's change, including zeros and drops.
Keep item pages separate Supabase and Convex stay on their item pages. The scenario composite uses only the Web frontend board weights for ordering and home cards; it does not replace the raw ranks in the fused table.