Nate Herk | AI Automation

I Tested Opus 5.5 at Every Effort Level. What You Need to Know.

Sep 24, 2026 27 min
claude opus 5.5ai agentsllm benchmarksai automationweb development
Watch on YouTube Follow Nate Herk | AI Automation on Rundown — free

Summary

AI summaries can be incomplete or wrong. Verify anything important against the original video.

Nate Herk tests the AI model Claude Opus 5.5 across all effort settings, from low to 'ultracode,' to evaluate performance, costs, and output quality when building a 3D virtual conference world.

This video provides a detailed technical evaluation of Claude Opus 5.5 by running identical, complex tasks across all six available effort levels (Low, Medium, High, Extra, Max, and Ultracode). Nate Herk tasks the AI with generating a fully functional, walkable 3D environment based on provided video assets of an AI tech conference. He methodically tracks and compares critical performance metrics for each run: total run time, total API costs, token usage, the number of automated verification checks triggered, and questions asked by the model during the process. The results reveal clear trade-offs, with higher effort levels generally yielding more polished and feature-rich environments but significantly increasing time and cost. The author identifies 'Extra' as his top-performing choice, balancing technical capability with efficiency, while noting performance bugs in 'Ultracode'.

Verdict

Claude Opus 5.5
ai model · Varies by effort level ($3.91 to $50.38)

Extra effort provides the best balance of speed, cost, and output quality for complex task automation, while 'Max' provides diminishing returns for the significantly increased cost.

Depends

Pros

  • ✓ Higher effort settings enable complex 3D world creation 0:44
  • ✓ Extra effort is cost-efficient relative to capability 25:36

Cons

  • ✗ Higher effort tiers significantly increase latency 23:13
  • ✗ Ultracode setting exhibits buggy, inconsistent behavior 22:06

Specs

Max cost multiplier vs low 12.9x 22:50
Max check multiplier vs low 2.3x 22:50

Best for

  • AI automation developers
  • power users optimizing API spend

Not for

  • users requiring simple, fast queries

Key Points

  • 0:44 The prompt asks the AI to build a 3D, walkable virtual tech conference world using video recordings as assets.
  • 2:27 Low effort results: 16m 43s run time, $3.91 cost, 22 checks, 0 questions asked.
  • 7:01 Medium effort results: 1h 13m run time, $12.44 cost, 23 checks, 0 questions asked.
  • 12:16 High effort results: 1h 7m run time, $16.31 cost, 22 checks, 1 question asked.
  • 17:53 Ultracode effort results: 1h 35m run time, $18.69 cost, 42 checks, 0 questions asked. Noted as feeling buggy.
  • 18:38 Extra effort results: 1h 31m run time, $25.92 cost, 34 checks, 0 questions asked. Labeled as the top performer.
  • 22:48 Final comparison of data: Max cost is 12.9x higher than low cost, and checks are 2.3x higher.
  • 23:15 Max effort results: 2h 28m run time, $50.38 cost, 51 checks, 0 questions asked.

Worth watching if: You are a developer or AI power user looking to optimize your Claude Opus 5.5 usage, as the video clearly quantifies the performance and cost trade-offs of different effort settings.

Get every Nate Herk | AI Automation video extracted like this

One daily email with structured extracts of every channel you follow. Free tier covers 15 videos a month.

Sign in with Google

No credit card. Free tier forever.

Watch on YouTube