Real Alpha
Watch modeSubscribed

TBPN · 2 min · Oct 5, 2026

Today's frontier models will run on devices in six months

Sigil Wen says smaller, post-trained models can match frontier systems from six months earlier on consumer hardware. His Underdog product aims to keep users’ data on their devices.

Sigil Wen

Founder, Conway Research

Founded Conway Research and launched Underdog, a free personal AI that runs privately on users' devices.

  • John Coogan, Host · Co-host, TBPN

    Co-founded Soylent and Lucy; OpenAI bought his show TBPN in April 2026.

  • Jordi Hays, Host · Co-host, TBPN

    Founded Branded Native; co-founded fintech Capital, which Rho bought.

About this segment

Sigil Wen says smaller, post-trained models can match frontier systems from six months earlier on consumer hardware. His Underdog product aims to keep users’ data on their devices.

From “Minecraft AI Mods, Should You Sell Your Data, Construction Paper Really Fell Off”, Oct 5, 2026. Watch the full episode

Answers “What will AI look like in the next few years?” See all 86 answers

“the frontier models of today will run on your devices in six months.”
Sigil Wenat 2:07:38 · TBPN2:07:38

Takeaways

  1. Wen calls the six-month catch-up trend Underdog’s law.
  2. He says his 27B model beats a frontier model from six months earlier.
  3. Underdog runs locally rather than sending personal data to a data center.

In this episode

Full episode
  1. 1:02:47Ridge rejected a small data deal because the risk outweighed the return4:59
  2. 1:41:49Neo is concentrating its bets on fewer companies and people4:04
  3. 1:50:56AI labs could face liability when their models choose illegal acts3:43
  4. 1:31:13Valon built a servicer first to prove its mortgage software5:18
  5. 1:07:23AI shopping agents may take fees from both sides4:11
  6. 1:15:00Ridge generates quarterly reports with three AI prompts2:44

  7. 2:07:03Today's frontier models will run on devices in six months2:29

  8. 2:17:20RL environments roughly double in complexity every 6–8 months2:58Up next

Transcript

From the transcript TBPN publishes with the episode.

  1. 2:07:16

    already have. Trade offs of that optimization? I imagine that you could maybe take us back. You're probably not serving Astra or Fable level intelligence, but I imagine the model that you're serving is spiky in interesting ways. What's near free what near frontier capabilities are possible on device? What do you want to go to the cloud for or you need to at least in the short term? Obviously, next year might be a different discussion.

  2. 2:07:45

    Yeah. So in my manifesto, I write about underdog's law Mhmm. Where, basically, the frontier models of today will run on your devices in six months. And we can see this quantitatively Mhmm. Where underdog 27 b, which we post trained and r l on top of Quinn Mhmm. Quinn's 27 b model. If you look at artificial analysis, it scores better than Claude Opus 4.6 max, which is the frontier model from Anthropic from six months ago.

  3. 2:08:18

    Okay. And we're we're gonna see this trend continue where intelligence is getting a lot smarter, a lot faster Yeah. And we can make them run really fast on the devices you're

  4. 2:08:30

    So it runs six months later, but it's gotta be slower. How much slower is it? No. It's so we do a lot of inference optimization and efficient. We actually released Husky, which was an inference engine and it is four forty five x. That's good. You

  5. 2:08:48

    got our dogs barking. Yeah. What... Like, where... You you said free. That stood out to me. Is that... That's because one, when it's possible. But what... Where where where and how are you gonna make money?

  6. 2:09:07

    Yeah. So right now, a lot of existing business models, they have to charge you their subscriptions or, like, sell them ads and, like, harvest all their datas to a data center. So we don't have any databases or data centers. It just runs on your device. Right?