← Anthropic
Wes Roth
Anthropic is about to IPO at a TRILLION DOLLARS
⏱ 16 min video · 3 min read2 Jun 2026
TL;DR
Wes Roth covers three major AI stories: Anthropic filing confidentially for a US IPO (potentially valuing it near $1 trillion), Claude Opus 4.8 setting a new state-of-the-art on ARC AGI 3 and showing qualitatively different reasoning behavior, and rumors of GPT-5.6 dropping imminently with major coding and agentic leaps. The IPO is framed as the first real financial reckoning for the AI boom.
Key points
1
Anthropic has filed confidentially for a US IPO at a valuation of approximately $965 billion, which will force public disclosure of revenue, margins, inference costs, and enterprise retention data for the first time.
2
Claude Opus 4.8 achieved 1.5% on ARC AGI 3 — still low in absolute terms but far ahead of all other models — and observers noted it reasoned at a higher abstraction level, modeling objects rather than treating inputs as raw pixels.
3
On the Deep Suite software engineering benchmark (113 contamination-free tasks, 91 repos, 5 languages), GPT-5.5 still outperforms Claude Opus 4.8, though the Ultra Code effort mode was not tested.
4
Rumors point to GPT-5.6 and GPT-5.6 Pro appearing in OpenAI Codex backlogs, with reported improvements in coding, agentic abilities, and a potential 1.5 million token context window — possibly announced at an imminent livestream.
5
Wes Roth built a full autonomous economic simulation game using Claude Opus 4.8 in Ultra Code mode, which he is using as an independent LLM benchmark alongside up to 10 custom benchmarks he is developing.
Key arguments
→
Watch the Anthropic IPO prospectus closely once public — it will be the first audited financial data for a frontier AI lab and a genuine test of whether AI valuations are justified.
→
When evaluating AI model benchmarks, prioritize contamination-free, task-diverse evaluations like Deep Suite and ARC AGI 3 over leaderboard scores that may reflect training data memorization.
→
The rapid release cadence (GPT-5.5 in April, rumored GPT-5.6 in June 2026) signals that frontier model launches are shifting to rolling updates rather than annual events — plan AI tool evaluations accordingly.
Notable quotes
“This might be the biggest reality check yet for this AI economy.”
“It's not high. It's not extra high. It's not maximum. This is beyond maximum. This is ultra code.”
“We might have for the first time in history three trillion-dollar IPOs all in the same year — SpaceX, Anthropic, and OpenAI.”
Worth watching?
⏭️
Worth watching the full video?
The key details, numbers, and arguments are all captured here — skip the video unless you want to see the live demo of the autonomous economy simulation Wes built with Claude Opus 4.8.
Saved you some time? The creator still deserves a like.
Want this for your own channels?
Add the channels you follow. Every new video summarised and in your inbox the moment it drops. From £4/month.
Try it free