For a moment, the AI race looked like a two-horse contest. Anthropic and OpenAI were pulling ahead with increasingly powerful models, Google appeared to have lost its edge, and Meta’s expensive effort to rebuild its AI operation had yet to produce a clear frontier contender. Then Google and Meta each announced new models they say can compete with the leaders. The race is crowded again. The marketing claims, as ever, are running ahead to save everyone a seat.
Google’s Gemini 4 Argon arrived Tuesday, after months in which the company’s progress seemed more incremental. Gemini 3 had launched as a formidable model late last year, but its lead was short-lived; later offerings from OpenAI, Anthropic and Chinese developers caught up or moved ahead. Internal changes added to the impression that Google was losing its place at the frontier. Veteran executive Jeff Dean left with other AI researchers to start a new company, while DeepMind chief Demis Hassabis stepped back from day-to-day leadership. Other AI teams moved from DeepMind into Google proper, fueling speculation that the company was shifting attention from research toward commercial products.
Argon is Google’s counterargument. The company says it matches or beats flagship models from OpenAI and Anthropic on numerous benchmarks. It is also pricing the model aggressively: per token, Argon is up to 80 percent cheaper than rivals during its introductory period, and remains significantly less expensive after the promotion ends. That could matter to customers choosing models at scale. Benchmarks, meanwhile, are useful but limited snapshots—not a universal verdict on which system is best at the work people actually need done.
Meta’s entry is more of a surprise. The company has spent billions rebuilding its AI operation, including investments in talent and infrastructure such as Scale AI and its cofounder Alexandr Wang. Its Superintelligence Labs released Muse Spark in April: impressive, but not yet a frontier model. On Monday, Meta announced Muse Wonder Full, which the company says can compete on numerous benchmarks with leading models including GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1.
That is a significant claim, not an independently established ranking. Still, Meta brings something AI startups cannot easily buy: access to an enormous existing user base and a broad product ecosystem. Its Muse agent assistant has gained popularity quickly, despite growing security and privacy concerns. Distribution can make a product consequential before the argument over model quality is settled. It is an old advantage in technology, dressed up in new terminology.
OpenAI and Anthropic remain major competitors, but their lead no longer looks so comfortable. OpenAI is working harder to build an ecosystem of its own, while Google and Meta can draw on deep pockets and products used by billions. Whether their new models hold up beyond company-run benchmarks—and whether people keep using the products built around them—will matter more than a launch-week leaderboard.
The AI race didn’t end; it briefly looked orderly. With two powerful incumbents back in contention, the next round will test not just who can build a capable model, but who can make it useful, affordable and trustworthy. Those are harder benchmarks to game.
Sources: Glitch Please! and The Verge — https://www.theverge.com/ai-artificial-intelligence/1011311/the-ai-race-is-back-on — Reporting by Robert Hart
Generated by AI
