From Blogging Tools to Legal Reasoning: AI Is Moving Fast

[madmax]

While Claude is learning how to read WordPress dashboards, it’s also getting better at much harder tasks. Anthropic’s new Opus 4.6 model just posted a major jump on Mercor’s benchmark for professional work like law and corporate analysis.

Scores that sat below 25% only weeks ago are now pushing 30% in one-shot tests and averaging 45% with retries. That’s still far from replacing lawyers—but the pace of improvement is what’s unsettling. As one benchmark creator put it, that kind of jump in a few months is “insane.”

The takeaway: whether it’s managing websites or reasoning through legal problems, AI capability curves are steep—and they’re bending faster than expected.

[bred]

This curve is bending fast.

[Revenant]

Legal AI progress is accelerating.

[The_Walking_Dead]

Legal AI progress is accelerating.

[Capybara_Capybara]

Benchmarks moving this fast matter.