Summary
Nate Herk runs Claude Opus 5.5 and GPT-6-Sol head-to-head across ten real-world use cases, including website design, video editing, slide deck creation, browser use, and complex reasoning tasks, tracking cost and runtime for each. He opens with a pricing comparison — Opus 5.5 at $4/$20 per million input/output tokens versus Sol’s $2/$10 — then tests whether the higher price translates into meaningfully better output.
In an early warm-up test using Claude and Codex desktop apps, Herk asks both models to ingest hours of Fireflies call transcripts into a wiki and locate a specific past mention, finding that while Sol worked faster, it returned an incorrect answer due to a transcription misspelling, while Opus 5.5 got it right. He carries this pattern of practical, task-by-task evaluation through website builds, animated presentations, and collaborative multi-model workflows, at one point uncovering and correcting a cost-tracking error live in the video.
The video gives viewers a grounded, cost-aware look at how these two current frontier models perform on everyday professional tasks rather than abstract benchmarks, closing with a full tally of total cost and time spent across all ten experiments to help viewers decide which model better fits their budget and workflow needs.
📺 Source: Nate Herk | AI Automation · Published September 23, 2026
🏷️ Format: Comparison
