May 2026 was, by the numbers, Anthropic's month. Across 21,544 AI articles from 582 sources, Anthropic was the single most-mentioned company (1,642 stories) — ahead of OpenAI (1,127) and Google (927). The story behind that lead was its product: Claude was the most-covered AI product on the internet this month with 1,572 mentions, nearly triple ChatGPT (559), driven overwhelmingly by developer adoption of Claude Code (384) and the launch of Opus 4.8. Reports of a roughly $930B valuation and previews of the restricted "Mythos" model only sharpened the narrative that the frontier race is now genuinely three-way — and that Anthropic is setting the pace on coding and agents.

The frontier model race tightened

On the model side, Gemini led coverage (361) as Google pushed its 3.5 line and Android XR ambitions, but the more telling signal was the breadth beneath it. Open-weight models punched far above their weight: Meta's Llama (111), Alibaba's Qwen (99), DeepSeek (89) and Google's Gemma 4 (84) all out-charted several proprietary flagships. A recurring May theme — "5 open-source models you can run locally and never pay for again" — reflected real momentum toward self-hosted, privacy-preserving AI. Meanwhile GPT-5.5 (90) and the back-to-back Opus 4.8 / Opus 4.7 releases kept the closed frontier moving.

Agents and infrastructure, not hype

The month's technical discourse was strikingly practical. The most-discussed engineering topics weren't breakthroughs but production problems: multi-agent orchestration, persistent agent memory, test-time compute ("thinking longer, not training bigger"), KV-cache optimization, and the governance of shipping AI products at scale. Tools like Codex, GitHub Copilot and Cursor anchored a clear shift: the conversation moved from "what can models do" to "how do we run them reliably and cheaply." On hardware, Apple's M5 silicon, AMD's Strix Halo, and Cerebras WSE all surfaced as the infrastructure layer for local and edge inference.

The people driving the story

Elon Musk dominated mentions (127) — though as much for a bruising week in court as for xAI — followed by OpenAI's Sam Altman (87) and, notably, Andrej Karpathy (52), whose commentary on AI engineering kept resonating with developers. NVIDIA's Jensen Huang (28) and DeepMind's Demis Hassabis (18) rounded out the field, the latter increasingly tied to AI-for-science and drug discovery.

The throughline of May: the AI industry matured past the demo. Coverage skewed overwhelmingly toward media analysis (17,986 articles) and community discussion (3,052) rather than raw research (506) — a field now spending more energy on deployment, cost, governance and trust than on chasing the next benchmark. The winners were the ones making AI dependable in production.