[{"data":1,"prerenderedAt":14},["ShallowReactive",2],{"artikel-en-berita-metr-anthropic-kemajuan-ai-dwi-kali-9-november-2026":3},{"slug":4,"lang":5,"title":6,"description":7,"date":8,"updated":9,"topic":10,"product":9,"cover":11,"readMin":12,"html":13},"berita-metr-anthropic-kemajuan-ai-dwi-kali-9-november-2026","en","METR: Anthropic's models are improving twice as fast, driven by AI itself","An independent audit by METR found Anthropic is improving its models' capabilities roughly twice as fast as before, with most of the acceleration coming from using AI to do AI research itself.","2026-11-09",null,"berita","\u002Fimages\u002Fartikel\u002Fberita-metr-anthropic-kemajuan-ai-dwi-kali-9-november-2026.webp",4,"\u003Cp>METR, an independent AI safety research organization, published a report on 5 November 2026 on how far AI is accelerating AI development inside Anthropic. METR staff were embedded at the company from 12 August to 29 October 2026 with access to internal data. The headline finding is striking: Claude&#39;s capabilities are now improving roughly twice as fast as they were before October 2025, and the main driver is not bigger computers but AI being used to do the research itself.\u003C\u002Fp>\n\u003Ch2>1. What METR measured\u003C\u002Fh2>\n\u003Cp>METR was not an outside observer guessing from press coverage. The team received unprecedented access to training metrics, internal benchmark scores, AI speedup analyses and interviews with Anthropic researchers. The assessment lasted 11 weeks.\u003C\u002Fp>\n\u003Cp>The central figure: as of October 2026, Anthropic is likely improving its models&#39; capabilities on well-specified tasks at around twice the rate recorded between March 2024 and October 2025. METR flags its own uncertainty: the true acceleration could be as low as 1.3x or as high as 3x the historical rate.\u003C\u002Fp>\n\u003Ch2>2. Two years of progress in one year\u003C\u002Fh2>\n\u003Cp>For benchmark scores, METR used AECI, its version of the Epoch Capabilities Index. Since October 2025, Anthropic&#39;s benchmark scores have grown at an estimated 1.9x the pre-October 2025 rate, with a 95% confidence interval of 1.5x to 2.8x. Put simply, the last year of work amounts to roughly two years of historical progress at the previous pace.\u003C\u002Fp>\n\u003Cfigure>\n  \u003Cimg src=\"\u002Fimages\u002Fartikel\u002Fberita-metr-anthropic-kemajuan-ai-dwi-kali-9-november-2026\u002Fmetr-fig1.webp\" alt=\"METR's AECI chart showing Anthropic benchmark scores rising more steeply after October 2025\" loading=\"lazy\" decoding=\"async\" width=\"950\" height=\"789\">\n  \u003Cfigcaption>METR's AECI chart shows Anthropic's benchmark scores climbing more steeply since October 2025. Image: METR.\u003C\u002Ffigcaption>\n\u003C\u002Ffigure>\u003Cp>METR also states it is confident the acceleration was not simply caused by scaling up training compute. Historically, more compute has been the main driver of AI progress. This time, the researchers ruled it out as the primary cause.\u003C\u002Fp>\n\u003Ch2>3. AI accelerating AI\u003C\u002Fh2>\n\u003Cp>The main cause of the acceleration, according to METR, is Anthropic&#39;s extensive use of AI to do its own R&amp;D. Interviews and internal surveys point to sizable productivity gains among staff.\u003C\u002Fp>\n\u003Cp>Humans still control the process, though. Most of the uplift comes from researchers using AI as a research assistant. METR notes there are &quot;some signs that fully autonomous contributions will play a larger role soon.&quot; That connects to recursive self-improvement, the idea of an AI system that can develop its own successor, a topic Anthropic also flagged in its IPO filings.\u003C\u002Fp>\n\u003Cfigure>\n  \u003Cimg src=\"\u002Fimages\u002Fartikel\u002Fberita-metr-anthropic-kemajuan-ai-dwi-kali-9-november-2026\u002Fmetr-fig2.webp\" alt=\"METR's self-reported uplift chart showing Anthropic teams reporting productivity gains from AI use\" loading=\"lazy\" decoding=\"async\" width=\"950\" height=\"594\">\n  \u003Cfigcaption>Self-reported speedup estimates from Anthropic teams working with AI assistance. Image: METR.\u003C\u002Ffigcaption>\n\u003C\u002Ffigure>\u003Ch2>4. No bottleneck in the next 12 months\u003C\u002Fh2>\n\u003Cp>METR reports it saw no evidence of significant bottlenecks that would halt capability progress in the next 12 months, whether from algorithmic progress, training compute, experiment compute or data. That means the accelerated pace could continue for at least another year.\u003C\u002Fp>\n\u003Ch2>5. Why the report is raising concerns\u003C\u002Fh2>\n\u003Cp>Two things make this report significant. First, it is the clearest evidence yet that AI is accelerating the development of AI itself, something that until recently was mostly theory. Second, it lands while Anthropic prepares for its IPO, and just after CEO Dario Amodei publicly called on all frontier labs to slow down model development.\u003C\u002Fp>\n\u003Cp>One detail drew attention too: Anthropic held rights to redact the report and used them extensively. METR says a majority of the evidence it found most informative was redacted, on grounds of removing confidential information that could accelerate other AI developers. As a result, the public has to lean on METR&#39;s judgment more than would be ideal.\u003C\u002Fp>\n\u003Ch2>What it means for you\u003C\u002Fh2>\n\u003Cp>The pace of AI development now looks set to continue or speed up, which means tools like Claude will get more capable faster than many people expect. For everyday users and business owners:\u003C\u002Fp>\n\u003Col>\n\u003Cli>\u003Cstrong>Do not delay adoption.\u003C\u002Fstrong> If today&#39;s models are already good enough for your work, next year&#39;s will be far better. AI skills are worth more if you start building them now.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Directing AI becomes the key skill.\u003C\u002Fstrong> As models handle longer, more complex work, the value of people who can brief AI precisely goes up.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Plan for oversight too.\u003C\u002Fstrong> Regulators, including Malaysia&#39;s, are drafting AI governance rules. Businesses using AI need human supervision in the loop, not blind delegation.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Trust independent audits over marketing.\u003C\u002Fstrong> Figures like these come from an external assessment with internal access, not a company press release. That is the kind of source worth following.\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Ch2>Sources\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fmetr.org\u002Fblog\u002F2026-11-05-anthropic-ai-rd-acceleration-assessment\u002F\" target=\"_blank\" rel=\"noopener\">Independent Assessment of AI R&amp;D Acceleration at Anthropic in October 2026\u003C\u002Fa> (METR)\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fwww.theverge.com\u002Fai-artificial-intelligence\u002F1029126\u002Fanthropic-metr-report-accelerating-ai-development\" target=\"_blank\" rel=\"noopener\">Why a new report on Anthropic automating AI development is alarming researchers\u003C\u002Fa> (The Verge)\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Findianexpress.com\u002Farticle\u002Ftechnology\u002Fartificial-intelligence\u002Fanthropic-ai-speed-up-development-metr-report-key-findings-10955551\u002F\" target=\"_blank\" rel=\"noopener\">How is Anthropic using AI to speed up its own development? Key takeaways from new METR report\u003C\u002Fa> (The Indian Express)\u003C\u002Fli>\n\u003C\u002Ful>\n",1795068892982]