codexvsclaudearena
New run
Public Same-Prompt App Build

Public result: same product brief, Claude Code and Codex branches

A public GitHub comparison where the same competitive-intelligence app prompts produced separate Claude Code and Codex implementations.

May 26, 20261 min readLow reviewer burden
Product

Claude Code resolves the task as a narrower analysis workflow, while Codex resolves it as a broader dashboard and decision-support surface.

Model

Both outputs preserve the requested schema, but they expose different assumptions about how much surrounding product surface the prompt invited.

Workflow Outcome

This is useful source material because the prompt, branches, screenshots, and generated code are all public.

Same-prompt generated results

Each panel shows the generated output for the same prompt, links to the original code, and keeps source credit visible.

Same prompt

Build a mobile-friendly black-and-white competitive-intelligence dashboard with frontend and backend after applying the shared research prompt and application prompt.

Source reuse
No explicit license found

This public repository is useful for branch-level inspection, screenshots, and outbound links. Mirror only short excerpts unless permission or a license is added.

Claude Code

Claude Code result

Open result

A controlled analysis workflow with a streaming API route, prompt builder, service layer, parser, and focused input/result components.

Raw generated code
Claude Code API routeClaude prompt builderClaude sector form
Representative code
const analysis = await openaiService.analyzeSector(validation.value!, onProgress);
controller.enqueue(sendSSEEvent(encoder, "data", { payload: analysis }));
controller.enqueue(sendSSEEvent(encoder, "complete", {}));
Streaming route excerpt
Codex

Codex result

Open result

A broader dashboard-style implementation with typed request validation, analysis generation, and a richer decision-support shell.

Raw generated code
Codex API routeCodex analysis serviceCodex dashboard shell
Representative code
const parsed = sectorRequestSchema.safeParse(payload);
const analysis = await generateCompetitorAnalysis(parsed.data);
return NextResponse.json(analysis);
Route plus analysis handoff excerpt

Generated file structure

The same prompt produced different project boundaries. This is the first code-level difference visitors should see before reading individual files.

Claude Code
  • app/api/analyze/route.ts
  • components/SectorInputForm.tsx
  • lib/hooks/useAnalysisStream.ts
  • lib/services/claudeService.ts
  • lib/services/promptBuilder.ts
  • lib/services/responseParser.ts
Codex
  • app/api/analyze/route.ts
  • components/dashboard-shell.tsx
  • components/ui/textarea.tsx
  • lib/analysis.ts
  • lib/schema.ts
  • tailwind.config.ts
Gemini CLI
  • client/src/App.tsx
  • client/src/components/AnalysisDisplay.tsx
  • client/src/components/ui/index.tsx
  • server/src/index.ts
  • server/src/services/analyzer.ts
Systems and versions
Claude Code: Public branch: autoblocs/fgv-project@claude-code
Codex: Public branch: autoblocs/fgv-project@codex
Environment

Greenfield Next.js competitive-intelligence webapp generated from the same two public prompts in separate repository branches.

Prompt or task

Use the same context-setting research prompt and application-construction prompt to build a mobile-friendly black-and-white dashboard with frontend and backend.

Source credit

Original work: autoblocs/fgv-project.

AgentScope is not the original creator of this comparison. This page exists to make the public source easier to inspect and to point visitors back to the original repository, prompts, generated branches, and raw code.

What visitors can learn

This is a good example for the site because the same prompt does not produce the same product.

Claude Code leans toward a controlled workflow: input, analyze, view structured result. Codex leans toward a richer dashboard: request context, sample state, and decision-support cards. Both are plausible, but they imply different user experiences and review burdens.