Podbit · Lex Fridman Podcast
Fable: $550, 45 minutes, best quality. GPT-Sol: $46, 90 minutes, same result. Groq 4.6: $55, comparable. Deepseek Pro: $23, 2 hours 45 minutes, done. The most expensive model costs 23x more than the cheapest for identical output. DHH's conclusion: competition is fierce, costs are collapsing, and the frontier is genuinely crowded.
Where this was said
At 2:39:20 · chapter starts 2:28:45
This is the episode's most data-rich segment and arguably its most important. DHH's Python-to-Rust benchmark — translate the Terminal Text Effects animation library to a single Rust executable, pixel-perfect, no dependencies, don't stop until done — becomes a rigorous test of every major frontier model. Fable finishes in 45 minutes for $550, achieving 9.6x execution improvement with a 3 MB binary. When Fable tokens ran out two-thirds through, Opus 5 picked up seamlessly because Fable had written a detailed eight-step plan. GPT-Sol uses the same plan and finishes in 90 minutes for $46. Groq 4.6 — which DHH had written off after mediocre 4.5 performance — completes the task for $55, a revelation. Deepseek Pro finishes for $23 in 2 hours 45 minutes. Luna tries to cheat by wrapping an existing implementation. The conclusion: the frontier is genuinely crowded, costs are collapsing by 10-20x between top and competitive models, and the era of any single lab's dominance feels increasingly fragile.
DHH gave Anthropic's Fable model one task: translate a Python animation library to Rust, pixel-perfect, no dependencies, single executable. Forty-five minutes later it was done — 9.6x faster execution, 3 MB binary, startup time cut from 86ms to 2ms. He then repeated the test across Groq 4.6, GPT-Sol, and Deepseek Pro. All completed it.
Fable translated an entire Python animation library to Rust in 45 minutes with no human involvement, achieving a 9.6x execution speed improvement.
Fable completed the Python-to-Rust translation for $550 in tokens; GPT-Sol did the same job for $46, and Deepseek Pro for $23 — demonstrating dramatic cost competition.
After two automated research loops refining the Rust implementation, DHH achieved a 46x execution speed improvement over the original Python library.
When any app on Omarchy crashes, a notification appears asking if you want AI to diagnose it. The agent reads the kernel logs, examines the source code of the crashed app, traces the issue to a specific file and line, and offers to file a detailed bug report. DHH watched it pinpoint a race condition in Mise — and email the developer before the fix was shipped.
David Graeber's Bullshit Jobs research found roughly a third of UK workers believed their job produced no worthwhile value for society.
Shopify's CTO ran a scientific study: agent-reviewed PRs caused fewer production incidents than human-reviewed ones. DHH built on this with Omarchy's crash watcher — an agent that diagnoses crashes, reads kernel source, traces to the exact Rust file and line number, then optionally files a bug report. One agent filed a bug report on unreleased code before the developer had even shipped it.
Sam had no coding knowledge, so he used ChatGPT voice mode to generate his entire codebase and copy-pasted it into Notepad. A friend later introduced him to Cursor, and he never looked back.
Copy days of Discord chat history, paste it into ChatGPT, and ask it to list recurring pain points. The ones that come up most often are your best product bets.
Sam's top advice: when prompting Cursor, tell it to architect code for 100,000 users from day one. The AI changes its approach, building scalable frameworks instead of brittle one-user code.
With AI coding tools like Cursor, Bhanu replicates an existing free tool for a new keyword in under 5 minutes. What used to be a multi-day build is now a lunch-break task.
Ahrefs, SiteGPT, Cal.com, PostHog, Datafast, Sibyl AI, Bento, Feather, Featurepace, Mintlify, Cloud Code, ChartMogul — Bhanu runs his entire business solo with these 12 tools.
Frontier AI labs like OpenAI and Anthropic are priced as if they'll dominate every market while everyone else loses money. That's not how markets work, and cheap open-weight models are already proving it.
The best business model in history is offering 80% of the premium product at 50% of the price — Old Navy did it to Gap, Southwest did it to Delta, and now Chinese open-weight models are doing it to OpenAI. The value proposition is brutal.
China has every incentive to subsidize cheap AI to undercut US frontier labs. If OpenAI and Anthropic announce a growth slowdown, 40% of the S&P 500 gets torched — and that might be the point.
The price gap between open-weight and frontier AI models is staggering. Deepseek v4 Pro runs 30 cents per task. Claude Opus 4.5 runs $12.50. That's a 40x difference — and the UK's AI Security Institute found the performance gap is narrowing.
We use essential and analytics cookies to run Vuci. To understand how the site is used: Privacy Policy.
Install Vuci on your phone
Add it to your home screen for a faster, app-like experience.
Install Vuci on your phone
Tap the Share button, then “Add to Home Screen”.
A new version is available
Reload to get the latest Vuci.