All corrections
Substack February 25, 2026 at 10:32 PM

www.astralcodexten.com/p/moltbook-after-the-first-weekend

1 correction found

1
Claim
Remember, Claude 4.5 Opus has a four hour 50% time horizon.
Correction

METR’s published measurements put Claude Opus 4.5’s 50% time horizon at about 4 hours 49 minutes (and later estimates around 5+ hours), not 4 hours.

Full reasoning

Scott Alexander claims: “Claude 4.5 Opus has a four hour 50% time horizon.”

METR—the organization that introduced and publishes widely-cited “time horizon” estimates—has publicly reported a higher 50%-time-horizon for Claude Opus 4.5:

  • In a METR post summarizing their estimate, METR states Claude Opus 4.5 has “a 50%-time horizon of around 4 hrs 49 mins.” This directly contradicts the post’s “four hour” figure.
  • METR’s later “Time Horizon 1.1” writeup also reports the model’s 50% time horizon as 289 mins under TH1 (which equals 4 hours 49 minutes), and 320 under TH1.1 (which is 5 hours 20 minutes if interpreted as minutes, as shown in the same table formatting). Either way, METR’s published point estimates are not 4 hours.

Because METR’s published estimate explicitly says “4 hrs 49 mins” (and provides the underlying minute value), the statement “four hour” is not accurate as a factual numeric claim.

2 sources
Model: OPENAI_GPT_5 Prompt: v1.6.0