ARK Weekly Update: AI Agent to Drive Further Development at Anthropic and OpenAI, Grok 4.6 May Lower the Cost Curve of Cutting-Edge AI

BlockBeats News, August 23, Cathie Wood's ARK Invest released its weekly market report covering three main themes:

AI Agents Will Continue to Drive the Development of Anthropic and OpenAI: Anthropic submitted an S-1 draft to the SEC on June 1 in preparation for a potential IPO and has been engaging with investors to gauge market sentiment. As of the end of May, Anthropic's Annual Recurring Revenue (ARR) had reached $47 billion, up from around $9 billion at the beginning of 2026, a growth of over 5x in just 5 months; Data provider TickerTrends estimates its current ARR could exceed $74 billion. OpenAI's ARR is approximately $41 billion, double the $20 billion at the beginning of the year. The combined ARR of the two companies has exceeded $115 billion, surpassing the 12-month revenue of the combined SAP, Salesforce, and Adobe and approaching the $150 billion annualized run rate of Microsoft's Productivity and Business Processes segment. Both companies plan to fund large-scale AI infrastructure construction through public market financing.

Grok 4.6 May Further Drive Down the Leading Edge AI Cost Curve: SpaceXAI's Grok 4.6, designed for programming, agent tasks, and knowledge work, has key parameters of a 500k token context window, $2 per million tokens input, $6 per million tokens output, significantly lower than OpenAI's GPT-5.6 Sol ($5/$30) and Anthropic's Claude Fable 5 ($10/$50). Artificial Analysis assesses its Intelligence Index score at 61, on par with GPT-5.6 Sol, only lagging behind Claude Opus 5 and Fable 5 by around 1 to 2 points, but at a much lower price point. In terms of task costs, Grok 4.6 is around $0.84 per task, at the intelligence-cost Pareto frontier; AA-Briefcase's long-term agent knowledge work Elo score is 1577, essentially on par with Claude Fable 5's 1574. SpaceXAI has also introduced Grok Bot, expanding the competition from model performance to Agent software layer capable of long-running tasks. After model capacity improvements, cost measurement should shift from "price per token" to "price per completed task," and if training and inference costs continue to drop by 85% and 99.9% respectively each year, companies will deploy agents in more workflows, significantly accelerating AI adoption.

Minimal Residual Disease (MRD) Detection Proves Its Utility and Continues to Scale: Key evidence from bladder cancer trials IMvigor010 and IMvigor011, where ctDNA-positive patients saw a 50% improvement in median overall survival after receiving atezolizumab. Natera achieved 283,000 Signatera tests in the last quarter (a sequential increase of 34,000 tests), while all other solid tumor MRD tests combined only amounted to 41,000 tests, giving Natera approximately an 87% share of the solid tumor MRD market. Signatera is expected to reach around $1.5 billion in revenue in the fifth year, roughly double that of Cologuard during the same period. The market consensus opportunity is around $20 billion, with potential further expansion through cancer survivor monitoring, late-stage treatment monitoring, CAR-T, and global expansion.