Vibe coding 3D game dev: Claude Pro beats ChatGPT Plus on limits and code audits
AdvertisingBubbly546 · reddit · 2026-09-15
A developer building a Juno: New Origins-style 3D physics sandbox game compares subscriptions after heavy vibe coding on both:
- On ChatGPT Plus, GPT-6 Astra is fast and capable but tokens "vanish" — constant waits for the 5-hour window reset; handing Astra's code to GPT-5.6 Sol to continue went badly (fixes one bug, creates two).
- A Claude trial invite showed noticeably more usable headroom: fewer refresh waits, and Opus 5 Max audited the multi-file project and ironed out bugs better than GPT-6 Astra while apparently using fewer tokens.
- Claude Pro also costs about £2 less than ChatGPT Plus; ChatGPT still wins on app/interface and not needing separate API-style credits.
Verdict: for large vibe-coding projects, Claude Pro was the clearly better experience.
More from coding & agent
- 'Mom, shhh, I'm talking to Devin AI': the meme every dev relates to — marvinvonhagen · 2026-09-15
- Social Quack launches: buy-once Mac scheduler with MCP hookup to Claude and ChatGPT — ThePeterMick · 2026-09-15
- Agent fixes macOS-broken app autonomously, ships TestFlight build in 10 minutes — intellectronica · 2026-09-15
- Open-source tool adds SDLC workflow config and tracking on top of Orca multi-agent orchestration — Long_Philosopher3520 · 2026-09-15
- How to detect MCP capability drift when servers update tools: snapshot, diff and gate risky changes — daani_maas · 2026-09-15
- Brex CEO: Stop Building AI Agents, Build AI Employees and Sell the Work Itself — petergyang · 2026-09-15