Open and closed models are on different exponentials
Interconnects Nathan Lambert
A new take argues open and closed AI models aren't competing anymore, they're on totally separate growth curves. Closed labs will chase peak intelligence and premium prices; open models will win on cheap, wide diffusion across the economy.
Based on reporting by Interconnects, Nathan Lambert — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
There's a piece from Interconnects making the rounds that reframes the open-vs-closed AI debate, and it's a useful reset. The core claim: this isn't a race where one side catches the other. It's two different exponentials, and both can be true at once.
On the closed side, the argument leans hard on what's happening with coding agents right now, specifically the jump in capability once you cross the Opus 4.5 and Codex 5.2 threshold. People aren't switching to these tools out of laziness — they're switching because the output difference is undeniable for complex knowledge work. And once professionals feel that gap, they don't go back to "good enough." The author says they'd personally pay $2,000 a month for these tools today, which tells you where pricing power is heading. Anthropic and OpenAI, with Google likely closing in, are positioned as the only labs with the capital, talent and integration depth to keep pushing that frontier — think less "AI startup" and more Apple-meets-Microsoft, selling both a hard-to-replicate integrated product and high-leverage subscriptions across the economy. The piece even throws out a number: $2-10 trillion valuations for OpenAI and Anthropic within 5-10 years, forming something like a cloud-computing-style oligopoly.
The open side tells a completely different story, and it's arguably the more interesting one. Open models aren't trying to win the intelligence benchmark race — they're building an entire commodity stack. No single company owns the integration, so pricing gets pushed toward zero across a layered ecosystem: inference providers like Together, Fireworks and OpenRouter, finetuning platforms like Tinker and Prime Intellect, and the hyperscalers quietly running open weights alongside their frontier partnerships. Enterprises that don't need the absolute best model, just one that clears a
My take — AI-written commentary, not fact-checked reporting
This tracks with something I keep telling people: stop treating open vs closed as a knife fight and start treating it as two separate markets with different customers. The closed labs are basically building premium infrastructure for people whose time is worth more than the subscription fee, while open models are quietly becoming the plumbing for everything else. Betting against either side feels like betting against electricity or plumbing — dumb, because the economy needs both, just not from the same vendor.
Read more about this at: Interconnects