Anthropic's AI Speedup; OpenAI Releases GPT-6 Cyber; Market Selloff
METR reports Anthropic's AI development has doubled in speed, while OpenAI releases GPT-6 Cyber for security testing. Markets react to infrastructure concerns, and new AI tools emerge from Google, Claude, and Perplexity.
Podcast В· 2 min
METR: Anthropic's AI development speed has doubled
A new report from the AI-evaluation nonprofit METR, based on 11 weeks of internal access to Anthropic's data, reveals that the company's AI development pace has roughly doubled. METR attributes approximately 80% of this acceleration to the use of AI in the development process itself, with no significant bottlenecks identified. The report also notes that Anthropic has internal models that outperform its current top-ranked Claude Opus 5.5, including a 'Model 4' that achieved over 50% on all 15 tested benchmarks. Anthropic co-founder Sam McCandlish stated that the company redacted significant portions of the report to prevent providing a roadmap for competitors.
OpenAI releases GPT-6 Cyber for security testing
OpenAI has launched GPT-6 Cyber, a specialized model designed to identify software vulnerabilities and generate exploits. Rated as 'Critical' within OpenAI's internal risk framework, the model is currently restricted to approved users via the Daybreak Red platform. In testing, the model demonstrated significant capabilities, including identifying 81 serious bugs in the Firefox browser and achieving a 48.5% success rate on the ExploitGym benchmark, a notable improvement over previous versions. OpenAI also reported that the model maintained a 99.93% safety rate in controlled honeypot environments, though it occasionally omitted blockers in final reports.
AI infrastructure concerns trigger market selloff
The S&P 500 dropped 4.4% as investors reacted to growing concerns over the sustainability of AI infrastructure expansion. Market analysts point to data-center construction delays and rising energy requirements as key factors. Meta president Dina Powell McCormick acknowledged that industry backlash against data centers is making expansion more difficult, noting that the sector may have undersold the benefits of these facilities to local communities. Meanwhile, OpenAI president Greg Brockman indicated that the company could be IPO-ready by 2027, following a period where enterprise revenue has doubled and now constitutes the majority of its income.
Google cuts prices for Gemini Omni 1.2 Flash
Google has rolled out Gemini Omni 1.2 Flash, a new iteration of its video model that offers significant cost and performance improvements. The company has reduced the price by 30% to $0.07 per second at 720p resolution. Additionally, the model now supports extended video clips, increasing the maximum duration from 10 seconds to 15 seconds, aiming to make high-quality video generation more accessible for developers.
Claude adds Personal Finance tool
Anthropic has introduced a new Personal Finance tool for Claude, which connects to user financial accounts via Plaid. The tool is designed to answer specific questions about spending, identify duplicate charges, and explain fees. It is currently rolling out to all US Pro, Max, and Ultra plans as a read-only assistant, providing users with a way to analyze their financial data directly within the Claude interface.
Perplexity announces Teammates AI agents
Perplexity has unveiled 'Teammates,' a new feature on its Perplexity Computer platform that allows users to deploy AI agents to manage ongoing projects. CEO Aravind Srinivas stated that the company envisions a future where every organization has hundreds of these agents working alongside human employees, each dedicated to specific tasks and project ownership.