Claude

Our verdict on Claude

Claude is the model we trust most to turn a file or a one-line idea into something visual and usable, right in the chat window. Its Opus and Fable models have set the bar for code quality on the show since mid 2025. The tradeoffs are cost and speed: the best Claude models are slow and expensive enough that Jordan's daily coding driver in June 2026 was OpenAI Codex.

This page covers Claude the model and chat app. For the coding agent, see our Claude Code page.

What we built with Claude

• CFO dashboards from documents (September 2025): Jordan gave Claude Google's 57-page 10-K and asked for a dashboard "as if I'm a CFO". It returned an executive summary plus revenue, net income, segment expenses, free cash flow and risks. We use the same trick with QuickBooks exports (Episode 11, at 10:24).
• Cholesterol medication 3D visualizer (September 2025): a labeled, rotatable model of the liver and blood vessels where you toggle Crestor, Zetia and Repatha and watch LDL drop. Built entirely by chatting with Claude, moving from 2D to 3D and then adding autoplay demos.
• Claude-generated n8n workflow (October 2025): one prompt produced the JSON for a podcast outreach workflow that imported into n8n in seconds.
• Ryz Labs Autoblogger and LLM Math Roaster: Claude is one of the text models writing blog posts, and one of four models generating Lean proofs.
• Kalshi trading bot and ScreenEval (January 2026): Opus 4.5 helped build the bot's algorithm and powers every score in ScreenEval. Sam's answer when Jordan asked which model: "Opus 4.5 for everything."

How we use Claude

1. Upload the raw file. A CSV, a spreadsheet export or a PDF. Jordan's 10-K was 57 pages and he did not read it first.
2. Ask a simple, role-based question. "Make me a dashboard as if I'm a CFO after reading this document" was the whole prompt.
3. Treat the result as throwaway. Jordan uses a dashboard once and makes a new one next time instead of maintaining it.
4. Coach it toward the outcome. For the visualizer, Jordan started with "build me an animated web tool" and iterated: 2D, then 3D, then labels and autoplay.
5. Use it to scaffold other tools. Ask Claude for an n8n workflow and import the JSON, or use Claude in the browser to build quick demos (Episode 38, at 16:45).
6. Move to Claude Code when it becomes a real app. The chat is for one-off tools; multi-page builds go to the agent.
7. Match the effort level to the job. With Fable, Jordan started on the highest setting, saw his spend jump, and turned it down.

And what I mean by outperform is you can give Claude a CSV, a spreadsheet, or a document, a PDF document, and it'll build you a pretty beautiful dashboard that is pretty cohesive.

— Jordan Metzner, Episode 11

How our view of Claude's models changed

• July 2025: Jordan said the market had almost settled on ChatGPT for most things and Claude as the winner for code (Episode 4).
• January 2026: after a week with his developers, Jordan said everyone was enthralled with the quality of Opus 4.5.
• February 2026: Opus 4.6 was a step function over 4.5. Emblem's August Kiles called it "truly an inflection point" and said he rarely needs to adjust its code (Episode 31).
• June 2026: Fable 5. Our guest Emanuele Melis rebuilt a physics puzzle game from his childhood in minutes. Jordan found it accurate and the design output world class, but very slow (Episode 46).
• July 2026: Fable was pulled. Jordan said the current models were not nearly as good and he wanted it back for coding (Episode 49).
• September 2026: on Fable 5.1, Jordan could not see a difference on a small coding task. David Petrou said it fixed Fable 5's loquacious, oblique language (Episode 51).
• October 2026: Jordan said the new Opus model was performing quite well and was significantly cheaper than Fable (Episode 53).

I found it found some security errors and bugs in my applications. It was able to implement features and libraries successfully and predictably.

— Jordan Metzner on Fable 5, Episode 46

Where Claude falls short

• Cost surprises. Arun Kalaiselvan of Arya Health said a change in Anthropic connection times made one bill about six times what they expected (Episode 24). Jordan ran out of tokens the one time he ran Fable with sub-workers.
• Speed. Jordan's Fable habit was to start a task before a World Cup match and check back at half time.
• Confusing effort levels. Jordan counted something like ten settings and wanted plain tradeoffs in cost and time instead.
• Locked to its own tools. David Petrou noted Anthropic does not let you use a subscription outside Claude Code with other tools, while OpenAI has been more welcoming.
• Adoption depth. Jackson Reiter said real estate firms all love Claude but mostly use it for basic automation while the work stays in Excel (Episode 47).

Some teams make the cost work. Anis Bennaceur said Attention is an Anthropic shop that caps most staff at a few hundred dollars of extra spend, gives top people unlimited Claude, and relies on Claude's UI showing how much usage is left for the day (Episode 45).

Claude compared

• ChatGPT vs Claude: Claude won our dashboard and 3D visual tests; ChatGPT's output was wordier and needed more prompting.
• Claude vs Gemini: David Petrou argues cheaper, faster models like Gemini Flash may win on cost versus quality.
• Claude Code vs OpenAI Codex: the agent-level matchup, where Codex has taken over much of Jordan's coding.

Episodes featuring Claude

• Episode 11: the 10-K to CFO dashboard demo, and the ChatGPT comparison.
• Episode 13: the 3D cholesterol visualizer built in chat.
• Episode 17: Claude writes an n8n workflow.
• Episode 29: ScreenEval runs on Opus 4.5.
• Episode 46: first impressions of Fable 5.
• Episode 53: the new Opus, cheaper than Fable.

FAQ

How do you use Claude?

Our favorite use is to upload a CSV, spreadsheet or PDF and ask for a dashboard. Jordan turned Google's 57-page 10-K into a CFO-style dashboard with one simple prompt, and we do the same with QuickBooks exports.

Is Claude better than ChatGPT?

For visual output, in our tests, yes. When we gave both the same 10-K, ChatGPT's summary was too long and its charts needed extra prompting. Jordan also built a 3D medication visualizer in Claude and said ChatGPT would not have come out nearly as good.

Is Claude Fable worth it?

Jordan found Fable 5 accurate, with world-class design output and real security bug finds, but very slow and seemingly expensive. After it was pulled he said the models that replaced it were not nearly as good.

Is Claude expensive?

It can be. Arya Health saw a bill about six times higher than expected after a change in its Anthropic connection times. In October 2026 Jordan said the newest Opus model was performing well and significantly cheaper than Fable.

Can you build an app in Claude without coding?

Yes, for small tools. Jordan built an interactive 3D cholesterol medication model by chatting with Claude, with no code and no third-party tools. For bigger apps we move to Claude Code.