AI Trends — 20 August 2026

OpenAI's Astra Solves Unsolved Problems

OpenAI's internal successor model, known as Astra, has reportedly worked through ten previously unsolved problems in mathematics and theoretical computer science, publishing formally verified proofs on GitHub for roughly $2,000 in compute. It's one of the more striking signals yet that frontier models are starting to produce genuinely novel research output, not just synthesize existing knowledge.

User Milestones Keep Falling

ChatGPT crossed 1 billion active users on July 31, making it the fastest consumer software platform ever to reach that mark. Gemini followed close behind, passing 1 billion monthly active users on August 11. The two platforms are now effectively racing each other for default-assistant status on billions of devices.

Prices Keep Dropping

GPT-5.6 Luna API pricing dropped 80% to $0.20 per million tokens, continuing a year-long trend of steep cost declines that's making AI automation viable for a much wider range of businesses. xAI's Grok 4.6, launched August 12, now matches GPT-5.6 Sol Max on the Artificial Analysis Intelligence Index at the same $2/$6 per million token pricing as its predecessor.

New Models and Agents

Google shipped Gemini Robotics ER 2 alongside three new agent-focused Gemini models aimed at production-scale workflows. Zai's GLM-5.3 API also went live this week, built specifically for coding, defensive cybersecurity, and long-horizon agentic tasks — part of a broader shift toward models tuned for autonomous, multi-step work rather than single-turn chat.

Regulation and Transparency Catching Up

US regulators now require frontier model developers to submit new models for government review up to 30 days before public release. Meanwhile, Anthropic is introducing machine-readable watermarks for AI-generated content to meet the European Union's new AI transparency requirements, and OpenAI has rolled out additional safety features and parental controls aimed at reducing risk for younger, unsupervised users.

Infrastructure Under Strain

The competition for AI dominance is increasingly a competition over physical resources: chips, electricity, and data center capacity. Nvidia's H200 chips are beginning to reach China, Samsung has raised advanced chipmaking prices, and European data centers are being pushed farther from major cities in search of available power. Some operators are experimenting with piping water directly into server halls as a more efficient way to remove waste heat.

This newsletter was compiled from public reporting on AI developments over the past 48 hours.