OpenAI researcher says Claude is stronger, offers Anthropic six tips to beat OpenAI

On October 24, OpenAI researcher Will Depue posted a series of tweets openly offering advice to rival Anthropic, arguing Claude could beat OpenAI by doing six things: adding a usable goal mode, fixing the message queuing system, stopping over-refusal of harmless requests, stepping up hard technical optimization, and improving the product experience. He then clarified his position: 90% of his net worth is in OpenAI PPUs (profit participation units), so "I wish it weren't this way too — I'm trying," signaling he wasn't deliberately disparaging his employer.

Confirmed

• Will Depue listed six things Claude needs to do to beat OpenAI: a usable goal mode, fixing message queuing, stopping over-refusal of harmless requests, stronger hard technical optimization, better experience, and more
• He disclosed that 90% of his net worth is in OpenAI PPUs, distancing himself from any accusation of trashing his own company's products
• He posted side-by-side screenshots criticizing Codex's output presentation: the competitor's markdown viewer scrolls properly with comfortable formatting, while Codex crams content into a wall of text
• He believes many OpenAI researchers have no access to Claude and lack opportunities to work on hardcore technical projects, so they don't realize how much stronger Claude actually is
• He did credit Codex with strengths: good product feel, excellent computer use, and better queuing and messaging systems than Claude
• His advice: spend two days a month away from your day job actually using the competitor's product

Why it matters

• This is a rare case of an OpenAI researcher publicly admitting the competition is stronger on some dimensions, and the judgment carries extra weight coming from someone deeply financially tied to OpenAI (90% of net worth in PPUs)
• The post also lays out each side's weaknesses — Claude needs to fix queuing and over-refusal, Codex needs to fix its markdown rendering — offering concrete comparison points for watching the two products compete
• If the claim that "researchers can't access competitors' products" is true, frontier labs may suffer from internal information bubbles that skew product iteration decisions

2026-10-24 ~ 2026-10-24 · 5 related posts

• OpenAI Researchers Don't Know How Good Claude Is, Says Will DePue — willdepue · 2026-10-24
• [source] OpenAI researcher lists six things Anthropic must fix to crush OpenAI — willdepue · 2026-10-24
• [source] OpenAI researcher: 90% of my net worth is in OpenAI PPUs, and I still said it — willdepue · 2026-10-24
• Dev rips Codex's output rendering as rival markdown viewer shows it can be done — willdepue · 2026-10-24

1 near-duplicate retellings: willdepue