Report says AI is speeding up Anthropic's own model development up to 3x: recursive improvement is starting, not exploding

An independent evaluation firm, named in the source as Meter, reports that Anthropic is improving its models up to twice as fast thanks to AI-driven self-improvement. The estimated range runs from 1.3x to 3x. This comes from a short video report. We have not seen the underlying report or its method.

Some context helps. Recursive improvement means AI systems speeding up the research and engineering that produces the next AI systems. For years it was mostly a theoretical argument. An outside evaluator putting a number on it is a meaningful change, if the result holds up. It moves the debate from whether this happens to how fast.

The range deserves as much attention as the midpoint. At 1.3x, the speed-up is real but modest, similar to a strong productivity tool. At 3x, the pace changes. A spread that wide says the measurement is still uncertain. The note also does not say what "faster" measures. It could mean coding throughput, experiment cycles, or end-to-end model quality. We will not assume.

Our reading: this is the start of a feedback loop, not an intelligence explosion. Humans still set the goals, budgets and safety checks. The point is that an independent party is measuring the loop at all. That fits a pattern we keep seeing: outside verification of what labs claim matters more and more. It also strengthens our view that Anthropic and OpenAI lead the frontier. Faster internal iteration compounds that lead.

The implications cut both ways. In the short term, a faster development cycle puts pressure on safety evaluations and regulators to keep up, and it concentrates advantage in the labs that already lead. Over the long term, the same acceleration is what could bring forward the payoffs we care about: faster science, medical breakthroughs and greater abundance. Our view: welcome the measurement, demand the method, and pace governance to what is actually measured, not to the hype.