InstaDeep CEO: Anthropic's unreleased Models 3 and 4 sit halfway to AI2027's Agent-2
connoraxiotes · x · 2026-11-09
InstaDeep CEO Karim Beguir summarized where the AI frontier stands, charted against the AI2027 scenario (80% coding time horizon):
- Unreleased models crossed a key threshold: Anthropic's unreleased Models 3 and 4 now sit halfway between Agent-1 and Agent-2 on the AI2027 trajectory (graphs fixed for a bug flagged by the AI2027 authors).
- Public models lag massively: gap is roughly 1 day vs. 1 week of autonomous high-quality work, and it's widening fast, fueled by RSI.
- Automated coder on schedule: we're right on track with AI2027's prediction of an automated coder (AC) by April 2027 — "wild," in Beguir's words.
- Conclusion: if the trajectory holds, this means a world-changing intelligence explosion in 2027, and the public needs to know.
Related event: Unreleased models already at AI2027 "Agent-1.5" level, InstaDeep CEO says(2 posts)→
More from AGI Musings
- Anti-AI Arguments Are Arguments Against Intelligence Itself: A Four-Point Case — JHochderffer · 2026-11-09
- Making up numbers transparently beats hiding behind vibes, researcher argues in EA debate — davidmanheim · 2026-11-09
- Study: AI chats introduce a new vendor in two-thirds of software buying conversations — VeryWellVersed · 2026-11-09
- Hot take: good AI judgment is now the new median reviewer — 3scorciav · 2026-11-09
- AI insider warns public perception will worsen as frontier progress outpaces intuition — manosaie · 2026-11-09
- NYU paper: AI assistants push retail investors into index funds by 7.6pp — danielrock · 2026-11-09