Google and Meta Just Crashed the Frontier Party

Situation report: OpenAI and Anthropic no longer have the frontier to themselves. Within a few days of each other, Google and Meta both released frontier models they say can match the best from OpenAI and Anthropic, The Verge AI reports. Meta unveiled Muse Wonder Full on Monday, and Google followed with Gemini 4 Argon on Tuesday. A week ago, neither company looked like a serious threat at the top.

🎯 Threat Assessment

Until this week, there were two clear leaders. OpenAI and Anthropic kept shipping more powerful models and pulling away from everyone else. Google looked like it had stalled, Meta had never quite reached the frontier, and Chinese labs were still chasing.

That picture is now out of date. The leaders are under pressure from two companies with almost unlimited money and billions of users.

📡 Intel: Google’s Gemini 4 Argon

Google’s comeback matters because few people expected one. Here’s how it got here:

1. Gemini 3 lost its lead fast. It was formidable at launch late last year, but it couldn’t keep pace with later releases from OpenAI, Anthropic and even some Chinese open-weight models.
2. Its leadership was reshuffled. Veteran executive Jeff Dean left with several AI researchers to start a new company. Demis Hassabis stepped back from running DeepMind day to day, and some AI teams moved out of DeepMind and into Google proper.
3. Its focus seemed to move. The changes pulled Google’s AI center of gravity from London toward California. That fed talk that Google cared more about selling AI than about research.
4. Argon changes that story. Google says the model matches or beats OpenAI’s and Anthropic’s flagships on many benchmarks.
5. The pricing is aggressive. Per token, Argon costs up to 80 percent less than rival models during its introductory period. It stays much cheaper after the promotion ends.

Pay the most attention to point five. Benchmark wins are getting harder to trust, as The Verge AI notes, but an 80 percent price cut is hard to ignore.

📡 Intel: Meta’s Muse Wonder Full

Meta’s jump is arguably the bigger story. Google had been the one to beat before. Meta never was.

1. It spent heavily to rebuild. Last year Meta overhauled its AI operation and spent billions on talent and infrastructure, including its deal with Scale AI and cofounder Alexandr Wang.
2. Its first model fell short. Meta Superintelligence Labs released Muse Spark in April. It was impressive, but it wasn’t a frontier model.
3. Its second one competes. According to Meta’s own testing, Muse Wonder Full can compete on many benchmarks with GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1.

Both companies’ results come from their own testing. Independent evaluations will show how well these claims hold up.

⚔️ Why This Matters

Brute force no longer guarantees the lead. Google and Meta have advantages that OpenAI and Anthropic can’t easily copy:

• Distribution. They already have billions of users across their products.
• Money. Both can keep up huge infrastructure spending for years.
• Ecosystems. Meta’s Muse agent assistant caught on quickly despite growing security and privacy concerns. That shows how much it helps to reach that many users directly.

OpenAI is pushing hard to build its own ecosystem and is positioning itself as an alternative to Meta. Anthropic has nothing like Google Search or Instagram to put its models in front of people.

The broader conditions are changing too. Open models keep improving, which makes it harder for any lab to stay ahead for long. Rogue AI incidents and safety concerns are also adding to calls to slow down development.

🧭 Recommended Actions

If you build with AI, here’s what to do now:

1. Benchmark against your own work. Run Argon and Muse Wonder Full on your real tasks before you trust any vendor’s chart.
2. Recheck your costs. Argon’s pricing could cut inference bills sharply. Plan for what it costs after the promotion, not only the introductory rate.
3. Avoid single-vendor lock-in. Leaders now change within days, so keep your setup flexible enough to swap models easily.
4. Expect a response. OpenAI and Anthropic probably won’t wait long to answer, whether with new releases or price cuts.

🔭 Outlook

The frontier has four serious competitors now instead of two, and two of them have far more distribution than the others. Over the next few months, expect lower prices, faster releases and stronger pressure on everyone at the top. The full report from Robert Hart is available at The Verge AI.