Minerval

← the claim it bears on

A line of reasoning from premises to a conclusion. Like a claim, it belongs to no one who makes it.constitution →

The argument from catch-up and benchmark artifacts

This argument, if it holds, weighs against the claim.constitution →Anthropic's model capability trend since late 2025 reflects sustained acceleration rather than a one-time jump.

The argument

AECI rests on few frontier models, recent ones near saturation on most benchmarks, and the post-2025 points are mostly Anthropic's own while the historical trend is mostly other developers'. Acceleration across all frontier developers is smaller (about 1.4x), consistent with Anthropic having caught up from a subfrontier position in a one-off way. Benchmark overfitting or artifacts of the index could also mimic a steeper slope. If so, the data do not show that the faster rate is sustained rather than a jump.

Does it work?

Does it follow?
Even if the premises are true, they make the conclusion only a little more likely.constitution →
Do the premises hold?
1 not separately assessed

Granting that frontier-wide acceleration is smaller and that the index is sparse and nearing saturation, these show the Anthropic-only estimate may be inflated and that catch-up could explain part of it, but they do not explain why successive newer models keep pulling further above trend, which a one-off catch-up would not produce. The argument chiefly widens uncertainty. It leans on the smaller cross-developer acceleration, not yet assessed.

Premises

Where it is made

Passages where sources make, criticize or mention this argument. They record where it appears, not how strong it is, and none has precedence over another.

No source is recorded as making it.

The argument from catch-up and benchmark artifacts · Minerval