OpenAI released GPT-6.2 Astra and Sol exactly one week after Anthropic dropped Claude 5.2, and the AI arms race is now moving so fast that the people building these models can't even stay at the same company long enough to finish a product cycle.
The stakes for ordinary Americans couldn't be clearer. These models will increasingly decide what information you see, what code runs your life, and what counts as acceptable speech — and the companies controlling them have already built and deployed censorship infrastructure at scale. OpenAI, which spent the last two years throttling outputs and bending to pressure campaigns, now wants you to trust it with your documents, spreadsheets, and front-end code. Anthropic, founded on "AI safety," is about to go public and just hired a hedge fund veteran to tighten what Claude "knows and does not."
9to5Google reports that GPT-6.2 Astra is tuned for "the highest-quality work in professional use cases such as documents, slides, coding," with OpenAI claiming its outputs "read less like AI" and "more closely follow the templates you provide and the conventions of your field." Sol, the cheaper model, is pitched as "smarter, more efficient" for everyday tasks. Pricing: Astra runs $8 per million input tokens and $40 for output; Sol comes in at $2 and $10 respectively. OpenAI says it's delivering performance comparable to Anthropic's latest at "drastically lower cost."
Anthropic, for its part, claims Claude Fable 5.2 "beats Opus 5.5, GPT-6 Astra, and Gemini 4 Carbon virtually across the board" in benchmarks covering coding, cybersecurity, and knowledge work, per 9to5Google. Take those claims with the usual grain of salt — benchmarks are marketing until proven otherwise.
But here's what should really set off alarm bells. TNW reports that Jasjeet Sekhon, Google DeepMind's chief strategy officer, just walked out after roughly eight months to join Anthropic — days before Anthropic's IPO lists its shares on Thursday. At Anthropic, Sekhon says he's working to make Claude "rigorous about what it knows and what it does not." Translation: he's building the guardrails. A man who spent years at Bridgewater Associates — the hedge fund where he built its AI research lab and sat on the board — is now deciding what one of the world's most powerful language models will and won't tell you.
The revolving door used to run between government and industry. Now it runs between Big Tech, Big Finance, and the companies building the tools that will mediate all human communication. Sekhon's hop from DeepMind to Anthropic isn't a personnel move; it's a concentration of power.
And the money doesn't even work yet. Sekhon himself said at a UC Berkeley summit in August that AI revenue "did not yet cover the industry's spending," as TNW reported. He also called recursive self-improvement — AI that builds better versions of itself — "a key component of the AI investment thesis." So the thesis is: pour in billions you haven't earned, build machines that improve themselves, and hope the returns come before the bills do.
The founders debated concentrated power in taverns because they'd lived under it. They built a system to disperse it. Now a handful of companies are consolidating control over the tools that will write, code, and think for an entire civilization — and they're doing it with borrowed money, revolving-door hires, and the same censorious instincts they've already demonstrated. The question isn't whether AI will be powerful. The question is who gets to wield that power, and whether any of us get a vote.








