METR finds Anthropic's AI research speed roughly doubled

METR, an outside group that measures AI capabilities, ran a close-access assessment inside Anthropic from mid-August to late October 2026. They found that progress on well-specified research tasks is improving about twice as fast as it did between March 2024 and October 2025. Their internal growth metric rose roughly 1.9 times since October 2025, with a plausible range of 1.5 to 2.8 times. METR estimates that about 80 percent of this speedup comes from AI tools helping researchers work faster, not from simply throwing more training compute at the problem. The report also says it found no obvious bottleneck that would slow this trend down soon. At the same time, Anthropic staff interviewed for the assessment said AI still mostly assists with the hardest training work rather than doing it independently. The report is notable because it comes from an outside evaluator rather than Anthropic itself, giving it more weight as an independent check on how fast AI-assisted AI research is actually moving.

Why it matters

An independent measure of how much AI is already speeding up its own development, a key signal for forecasting future capability jumps.

Source: METR