The presenter separates provider claims and Artificial Analysis results from his own eight-task KingBench 3 testA benchmark is a standardized task or collection of tests used to compare AI systems under defined conditions.. Both models run at medium effortReasoning effort is the amount of internal computational work an AI model applies before producing an answer or action., but the video explicitly warns that effort labels are not equivalent across providers. Scores combine functionality, instruction following and subjective visual quality rather than representing a standardized measure of intelligence.
Opus performs better in the reported elevator, contact-lens-case and archery tasks, particularly where physical behavior and interaction matter. Sol ties on four tasks, including an SVG illustration, a combinatorics exercise, fine-tuningFine-tuning continues training a model on selected data so its behavior becomes better suited to a task, domain or operating environment. with a web interface and a working wristwatch. The creator reports 75/80 for Opus and 66/80 for Sol in this test set.
Four separate longer app buildsCode generation uses AI or another automated system to create source code from instructions, examples, schemas, or higher-level specifications. further favor Opus in the creator's judgment, although Sol produces a useful Blu-ray library and functional notes app. The conclusion weighs token pricingToken pricing is the rate an AI provider charges for processing input tokens, generating output tokens, or reading cached tokens. against the effort of repairing incomplete outputs. Repair cost was not measured, and the video contains inconsistent spoken context-limit figures, so those figures are not adopted here.
Watch on YouTube
