This week belonged to the models — and to the escalating OpenAI–Anthropic–Google race. The praise was loud, even giddy; Liberty’s Highlights even said Opus 4.5 delivered “more feel the AGI moments.”
We’re still waiting for December’s secondary prices to form our own view, but while we wait, we figured: if these models are so smart, let’s ask them what they think about valuations, multiples, and investor sentiment.
What the Models Think of the Models
This week delivered a double punch: Google and Anthropic both shipped major model upgrades. And inside OpenAI, the mood isn’t calm. According to The Information, Sam Altman warned colleagues that Google’s latest progress could “create some temporary economic headwinds” for the company.
Google’s gains came from pretraining — the foundation layer where a model absorbs the raw structure of the world from the open web and other broad sources. OpenAI has struggled here recently, and researchers widely agree: pretraining improvements drive generalization, the holy grail. If you want a model that can reason about things outside its training set, make scientific leaps, or help discover cures, this is the layer that matters.
Post-training, by contrast, sits on top. It teaches models to behave better — to code, solve math, follow instructions, sound helpful, or stay aligned with a specific domain. But post-training gains don’t automatically generalize. So when Google nails pretraining, it hits OpenAI where it hurts.
Altman told staff that OpenAI would regain momentum with a new model (codename Shallotpeat) and by doubling down on “very ambitious bets.” He framed the moment as a necessary dip: “We have built up enough strength as a company to weather great models shipping elsewhere. But staying focused on getting to superintelligence is critically important.”
Meanwhile, Anthropic dropped Claude Opus 4.5 — and it’s smarter, cleaner, and radically cheaper:
- Opus 4.1: $15 per million input tokens / $75 output
- Opus 4.5: $5 input / $25 output
It’s still pricier than GPT-5.1 or Gemini 3, but the 66% cost drop is massive. The standout isn’t just raw capability. It’s token efficiency. Opus 4.5 outperforms Sonnet 4.5 on SWE-Bench while using a fraction of the tokens. Anthropic also lifted usage caps, giving Opus users longer, more coherent sessions without memory collapse.
How did they make the economics work? Likely a blend of algorithmic improvements plus Blackwell GPUs ramping up for inference.
Ok, so what does it mean for us, investors?
I asked the three smartest: ChatGPT 5.1, Opus 4.5, and Gemini 3. Here are the takeaways.
ChatGPT 5.1
ChatGPT was the most dramatic. It expects a near-term re-rating: more enthusiasm and higher implied multiples for the “quality + efficiency” story (Anthropic) and a more cautious stance on OpenAI. It even predicted that Anthropic would cut prices.
Claude Opus 4.5
Claude was less dramatic on OpenAI, modest on Anthropic, and praised Google.
It also got hilariously confused about timelines — talking about OpenAI’s ~$150B valuation. Claude, really?
Still, its core point was solid: OpenAI’s forward revenue multiple likely compresses.
Claude’s overall framing:
- “The winner-take-all narrative is breaking down.”
- OpenAI faces the toughest position: defending leadership while others close the gap and undercut on cost.
- Anthropic has executed well but faces scaling challenges. Their cost structure depends on maintaining algorithmic efficiency leads, which requires continued R&D excellence.
- If Gemini is demonstrably better at the hardest reasoning tasks, Google can credibly claim technical leadership. The market may begin re-rating Google's AI optionality upward. Their AI capabilities have been embedded in a conglomerate discount; demonstrated leadership could unlock that value.
Gemini 3
Gemini was the most pleasant to deal with — structured, calm, neither too long nor too short, and free of embarrassing mistakes.
Its view: we’re hitting a major inflection point. The competitive game is shifting from a pure capability race to a battle of capability + efficiency + total cost of ownership (TCO).
Its predictions:
- OpenAI: Multiples compress. Slower growth and pricing pressure force investors to demand proof, not promises.
- Anthropic: Multiples expand. Its TCO advantage drives massive market share, and the lowest-cost producer wins volume when the market commoditizes.
- Google: Confirmed as the technological standard setter. Others follow; Google dictates the pace.
And now we wait for December’s secondary price updates for OpenAI and Anthropic — the real-world check on how quickly private markets are absorbing this shift.
Musk’s xAI Builds Its Own Power — Kind Of
xAI plans to build a small solar farm right next to its Colossus data center in Memphis. The company is working with a developer on an 88-acre installation, which should generate roughly 30 megawatts of electricity.
Sounds big — but it’s only about 10% of what Colossus is expected to consume.
Private Markets: Still Running Hot
Revolut closed a new share sale at a $75 billion valuation, cementing its status as one of Europe’s most valuable private tech companies. The company didn’t disclose the size of the raise but confirmed that employees were able to cash out as part of the deal.
SpaceX is preparing its next twice-annual tender offer, with pricing expected in December. The company was last valued at $400 billion this summer in a secondary sale, and insiders say demand remains intense while an IPO is still far off.
Kalshi jumped to an $11 billion valuation after raising a massive $1 billion round from Sequoia and CapitalG, underscoring investor appetite for prediction markets.
Physical Intelligence (Pi) raised $600 million at a $5 billion valuation (excluding the investment). Pi develops robotic AI models that learn from real-world experience; its latest model, π*0.6, shows faster performance in tasks like box assembly and folding laundry thanks to on-robot learning.