AGI at the Door: Claude Opus 4.6 Is the ChatGPT Moment for Agents

At the end of 2025, most frontier researchers in Silicon Valley estimated that artificial general intelligence (AGI) was five to ten years away. Four months later, consensus among researchers, founders, and practitioners has advanced significantly. AGI is now transitioning from a long-term hypothesis to a near-term problem.

A plausible timeline, in our view:

• 2026: SWE AGI - In most mainstream software engineering workflows, AI can independently decompose tasks, write code, test, debug, and submit pull requests; humans are only needed for goal-setting and final review.

• 2026–2027: White-collar AGI - In structured, digitized, tool-chain-clear knowledge work (Excel, PowerPoint, data wrangling, document processing, standard analysis), AI can handle all execution steps essentially.

• 2027–2028: AI self-iteration - AI significantly compresses the AI R&D cycle, systematically iterating on itself, approaching singularity.

Claude Opus 4.6 is the ChatGPT moment for agents. ChatGPT brought AI from the lab to the mainstream; Opus 4.6 was when the market realized AI could be given a job, run for hours, and deliver work directly into production. Mythos (Anthropic’s latest frontier model, several multiples of Opus in parameter count and compute) and Spud (OpenAI’s next-generation model, reportedly ~40% more capable than GPT-5.4) are the GPT-4 moment for agents, offering a significant increase in capability on a compressed timeline. Very similar to what we saw in early 2023, computational resources have again become the primary constraint: H100 and B200 rental prices have increased by 20%–22% over the past three months (Silicon Data), and companies such as Meta and Anthropic are securing multi-year cloud capacity contracts at prevailing rates.