METR report: Anthropic's AI models improving about twice as fast, driven largely by AI

Mixed · 2The AI safety nonprofit METR (Model Evaluation and Threat Research) published a report on Thursday 5 November 2026 finding that Anthropic's AI models are improving nearly twice as fast as before. 1 2 Centre · 1METR's 11-week assessment drew on confidential research, interviews and performance data from unreleased models, and found Anthropic's benchmark capabilities had been improving 1.9 times faster since October 2025 than previously. 1 Left · 1According to The Verge, benchmark scores suggest Anthropic made roughly two years of historical progress over the last year. 2

Mixed · 2METR said it was confident the acceleration was not due to faster scaling of training compute, and attributed most of it to AI itself; CNA reports METR put AI's share at perhaps 80 per cent but cautioned the figure was uncertain. 1 2 Left · 1Supporting evidence included Slack polls and interviews with Anthropic employees: staff using the most capable internal model self-reported being around 12 times as productive as they would be without AI, about double what was reported a few months earlier with Mythos 5. 2 Mixed · 2METR said AI is not yet capable of highly significant, fully autonomous contributions to the most important stages of model training, and that much of the acceleration comes from AI assisting human researchers, though there are signs autonomous contributions will play a larger role soon. 1 2

Mixed · 2The researchers found no convincing evidence of major bottlenecks or diminishing returns that could slow progress over the next year. 1 2 Left · 1The Verge reports that, per METR, Anthropic has finished training three unreleased models in the past six months, codenamed Model 2, Model 3 and Model 4, completing training in July, August-September and September-October respectively. 2 Left · 1Anthropic was allowed to redact the public version of the report. METR said it submitted a roughly 100-page full report and that Anthropic redacted a majority of the evidence METR found most informative, which it said means the public has to rely on its judgment more than would be ideal. 2

Left · 1METR also said Anthropic shared more information than initially scoped, but that its information pipeline was narrow and a small number of data requests were denied. 2 Left · 1Anthropic cofounder Sam McCandlish acknowledged on X that the company redacted a lot and that this makes METR's work harder to check, but said sharing what drives the speedup is commercially sensitive and risks giving others a roadmap; he said Anthropic will publish its own report on the speed of AI progress in the coming weeks. 2 Centre · 1The findings are seen as some of the clearest evidence yet that AI is accelerating its own development, raising the prospect of recursive self-improvement, which researchers have warned could be hard to align, monitor and control. 1

Left · 1The Verge reports the report prompted backlash among AI safety researchers; Apollo Research CEO Marius Hobbhahn said it suggests a pace faster than forecast in the AI 2027 scenario, though broadly in line with, perhaps slightly faster than, what many in the field expected. 2