This week's chart comes from METR, the independent evaluation group given access to Anthropic's internal research metrics between August and October. The orange line is the internal frontier, the models Anthropic uses inside the company. The green line is what the public sees, months behind. The dashed lines after today show the two things nobody can yet measure: how much of the acceleration comes from AI itself, and whether anything will slow it down.
Source: METR, Independent assessment of AI R&D acceleration at Anthropic, 5 November 2026. The chart is illustrative and not plotted from data.
METR's answer to the first question is cautious but clear. Anthropic is likely improving its models on well-specified tasks at about twice the rate it managed between March 2024 and October 2025. The 95% range runs from 1.5x to 2.8x. METR believes most of that gain comes from AI assisting researchers rather than from more compute, though it admits the attribution is uncertain and that much of the supporting evidence has been redacted. Staff polled on Slack said their most capable internal model made them about 12x more productive. The system as a whole moved at 2x.
The timing matters. Three days earlier, Anthropic amended its IPO prospectus to say it may imminently meet its own threshold for dramatically accelerated progress attributable to AI. The filing concedes that past accelerations were identified only after they had happened, because the indicators lag. In the same week, investors reportedly pushed the target valuation from $1.5 trillion towards $1 trillion, and the listing may slip.
Several researchers who had expected slower progress now talk about recursive self-improvement within months. METR's own report does not support that reading. It says AI cannot yet make fully autonomous contributions to the most important training work. But whilst Anthropic's financial fortunes are being shaped by politics and market forces, its technical process looks less constrained.