The Two-Horse Race Is Dead
For months, it looked like OpenAI and Anthropic had pulled away from the pack. Google seemed stuck in a rut of incremental updates, and Meta had yet to prove it could build a true frontier model.
Then everything changed in a single week. Google released Gemini 4 Argon. Meta unveiled Muse Wonder Full. And suddenly, the race is wide open again.
Google is back — and it’s undercutting everyone on price.
On October 13, Google made Gemini 4 Argon broadly available. The model matches or beats the best from OpenAI and Anthropic on key benchmarks — but the real shocker is the price.
Argon costs $2 per million input tokens and $10 per million output tokens, with a 95% discount on cached inputs. During its introductory period, it undercuts rivals by as much as 80%. It’s being called the “economical flagship” — a clear move to win cost-sensitive enterprise developers.
Meta finally built a model that belongs in the frontier conversation.
Meta unveiled Muse Wonder Full, a model it claims can go toe-to-toe with GPT-6 Astra, Claude Opus 5.5, and Claude Fable 5.1. It’s the first time Mark Zuckerberg’s company has truly arrived at the frontier.
Meta also released Muse Spark Open, an 816-billion-parameter open-weight model. It supports a 1-million-token context window and can run on two Mac Studios — bringing frontier-level capabilities to consumer hardware.
Anthropic is building a model that reads your mind, not just your prompt.
According to The Information, Anthropic is preparing to launch a new model that can infer the user’s underlying goal even when they don’t state it clearly. If you ask it to fix a bug, it might identify the entire class of problems causing the bug and ask if you want those fixed too.
The model hasn’t been named, and it’s unclear whether it belongs to the Fable or Opus line. But testers say it’s significantly better at reasoning and understanding intent than Anthropic’s current commercial models.
OpenAI’s next move is a mid-tier model — and it’s all about price.
OpenAI plans to release a new version of its Terra model in the coming weeks. Terra sits between the high-end Sol and the lower-cost Luna. The update is expected to improve capability while staying cheaper than GPT-6.1 Sol.
The strategy is clear: OpenAI isn’t just competing on raw intelligence anymore. It’s competing on cost, just like Google and Meta.
For a while, it felt like the AI race had two permanent leaders. This week proved that assumption wrong. Google and Meta didn’t just catch up — they changed the economics of the game.
The frontier is no longer a two-horse race. It’s a four-way street.
The Regulatory Reckoning
The model race isn’t the only thing heating up. This week, the legal and regulatory walls started closing in from every direction. Eight states sued OpenAI. The EU sent formal demands under the AI Act. And a leaked draft revealed that the industry’s biggest players are quietly trying to narrow the definition of “risk.”
Eight US states just sued OpenAI over harm to young users.
According to Reuters, attorneys general from eight US states have filed separate lawsuits against OpenAI. The suits allege that ChatGPT may cause harm to younger users and question the product’s design and safety mechanisms.
It’s one of the most coordinated legal actions against an AI company to date. The suits focus on product liability — not just what the model says, but how it was built and who it was built for.
The EU just sent formal demands to AI companies. This is the AI Act in action.
The European Commission confirmed that its AI Office has begun issuing formal requests for information under the EU AI Act. The requests are going to major AI developers, asking them to explain how their models comply with safety and transparency rules.
This isn’t a warning letter. It’s the first formal enforcement step. Companies that fail to respond or provide inadequate answers can face fines and restrictions on operating in the EU.
A leaked draft shows AI giants trying to narrow what counts as “risk.”
A secret draft document called the “Advanced AI Framework” has surfaced. It shows how major AI companies want to redefine risk controls.
Under the draft, only three categories would be severe enough to trigger a pause in development: biological weapons, model loss of control, and cyberattacks. Other risks — like job displacement, bias, and election manipulation — would be removed from the “pause” list. Developers would be required to report qualifying incidents within 72 hours.
Anthropic’s $2 trillion dream just got cut down to size.
At its pre-IPO investor day in San Francisco, Anthropic discussed a valuation of around $1.5 trillion — down from the $2 trillion target it had floated for its November IPO. The reason: revenue growth is slowing.
Anthropic reported Q3 revenue of $16.8 billion, up 45% from Q2. It also achieved adjusted profitability. But the deceleration is enough to make public-market investors nervous. The company also announced it will allocate 5% of its equity to a public interest fund.
The AI industry spent this week proving it can still build incredible things. Google shipped Argon. Meta shipped Wonder Full. But at the exact same time, the walls started closing in: lawsuits, EU enforcement, leaked drafts, and a valuation reality check.
The technology is accelerating. The accountability is finally catching up.
- Workday opened its HR and finance data to Claude — enterprise SaaS integration with AI just took another major step forward.
- Nvidia is in talks to deepen its investment in Reflection AI or acquire it entirely. Nvidia already invested $800 million in the open-source startup.
- Nvidia announced billions in new investments in Germany and bought stakes in French startups Doctolib and Genesis AI.
- A Dutch AI company, Intelic, is helping Ukraine’s air defense — and just secured a €30 million contract from the Dutch Ministry of Defense.
- Chinese model MiniMax M3.1 Flash launched as the latest in a wave of efficient, low-cost models coming out of China.
One ask this week: if this helped you understand something, forward it to one person who'd want to read it.
Just hit forward.
— THE AI CHAPTER Editorial Board