Comic #5·

Next: “One-shotted”

Benchmarks

1 / 4

Watch the animated strip on YouTube (opens in a new tab)

Benchmark v12 adds a question no human has solved, which we expect will hold for a full release cycle. The coastline remains unsaturated.

Read transcript

Panel 1

Two white-coated researchers stand in a seaside lab with an open laptop.

Researcher 1: “The models saturated it again.”

Researcher 1 looks at Researcher 2; Researcher 2 listens.

Panel 2

The same lab and laptop. Researcher 2 answers immediately.

Researcher 2: “Add harder questions.”

Both remain professionally composed.

Panel 3

A question card and escalating paper and chalkboard montage demand an unproven conjecture.

Both look at the work with closed mouths.

Panel 4

A result caption reads GPT6: 94.2%. Researcher 1 closes the laptop and walks into the calm sea.

The retreat is silent. Researcher 2 watches.