Quick answer
May 2026 snapshot: GPT-5 and Claude 4.7 are roughly tied at the frontier. Reasoning models are the new standard. AI agents are real but flaky. Multimodal is solved. AI video (Sora 2, Veo) is genuinely usable. Open source caught up (DeepSeek, Llama 4). EU AI Act is enforced. AI energy use is the new political flashpoint. Hype is fading; deployment is rising.
The mid-year is a good time to take stock. Six months ago we were arguing about whether GPT-5 would arrive. Now it has been out for a year and the conversation has moved on. Here is where AI actually stands in May 2026.
What got dramatically better in 2026
- Reasoning — o3, Claude Extended Thinking, Gemini Thinking transformed hard tasks
- Video generation — Sora 2 and Veo 3 are genuinely usable for short-form content
- Coding — Cursor, Windsurf, Claude Code are transforming developer workflows
- Open source — DeepSeek and Llama 4 are competitive with frontier proprietary models
- On-device AI — Apple Intelligence, Gemini Nano running locally on phones
- Voice and multimodal — Pi, ChatGPT Voice, Gemini Live are conversational at human level
What stalled or disappointed
- AI agents — autonomous web agents still flaky in real use
- AGI claims — no dramatic step change despite predictions
- AI replacing knowledge workers — happening, but slower than feared
- AI productivity boost economy-wide — measurable but smaller than hyped
- AI safety progress — outpaced by capability growth
What to watch in H2 2026
- GPT-6 and Claude 5 — both expected by year-end
- EU AI Act fines start landing — first enforcement actions coming
- AI energy crisis — data center build-out hits power grid limits
- Open source frontier — DeepSeek and Mistral pushing prices down
- AI in education — curriculum overhaul beginning at major universities
Honest take: 2025-2026 has been the "deployment" era — same model intelligence but radically more useful applications. The next breakthrough may come from agents finally working reliably.
Related reading
Bottom line
AI in mid-2026 is more useful than ever, slightly less hyped than a year ago, and quietly transforming how serious professionals work. The biggest gains came from deployment, not new capabilities.




