Summary
Nate Herk puts Claude Opus 5.5 through a practical test by giving the same prompt to every effort level, from low and medium through high, extra, max and ultra code. The task: build an explorable third-person 3D world that recreates a realistic tech conference from 105 GB of recordings from his virtual AIS Live event.
For each run, the video compares the finished result along with runtime, API cost, total tokens, the number of verification checks the agent performed, and how many clarifying questions it asked. The low-effort build is bright, glitchy and poorly branded, with people disappearing and video panels showing only still images. Higher levels add live chat, sound, seating, a working agenda, and video playback in the session rooms. One high-effort run took about an hour and seven minutes and would have cost roughly $16 at API rates.
The results do not follow a simple more-effort-is-better pattern, which is the main takeaway for viewers deciding how to configure Opus 5.5 for agentic coding projects. It is a useful reference for anyone weighing cost, speed and output quality when choosing an effort setting.
📺 Source: Nate Herk | AI Automation · Published September 24, 2026
🏷️ Format: Comparison
