The AI race is back on

The Verge argues the frontier race reopened this week. Google released Gemini 4 Argon, which it claims matches or beats OpenAI and Anthropic flagships on numerous benchmarks, priced up to 80% cheaper per token during an introductory period and still significantly cheaper after. Meta unveiled Muse Wonder Full, which its own testing says competes with GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1.

The piece notes Google had seemed to be slipping, with Jeff Dean leaving to start a company and Demis Hassabis stepping back from day-to-day DeepMind leadership, while Meta had spent billions — including on Scale AI and Alexandr Wang — without reaching the frontier until now.

We think the most important line in this story isn't about benchmarks at all. It's that Google and Meta have billions of users already in reach. A frontier model is a commodity faster than anyone expected; a distribution channel isn't.

That's good news for buyers. Four companies at the top instead of two means aggressive pricing — Google's introductory cut is the proof — and less dependence on any single lab's choices. Competition is how useful AI gets cheaper.

But competition also has a cost we've been tracking all month. The race is speeding up just as rogue-AI incidents and safety warnings pile up, and the piece itself notes calls to slow down are getting louder. More contenders with more money means more pressure to ship. That's exactly the dynamic safety researchers worry about.

A caution on the claims, too: both companies' frontier status rests largely on their own testing. Independent evaluations over the next few weeks will tell us whether this is parity or marketing.

I'm watching how OpenAI and Anthropic respond — price cuts, faster releases, or a renewed case that safety and trust are the real differentiator.

Google's Gemini 3 led briefly after its release late last year before rivals pulled ahead; Meta's Superintelligence Labs released Muse Spark in April.