Newer frontier models aren't always better; major labs have all shipped regressions
bindureddy · x · 2026-08-25
The author notes that a new model from a frontier lab doesn't guarantee improvement. Every major lab has officially shipped a regression with a "next gen" model: Anthropic's Opus 5 vs 4.8, OpenAI's GPT 4.5 vs 4o, and Google's Gemini 3.1 vs 3.5. It is easy to run in circles when chasing frontier intelligence.
More from Companies & People
- Anthropic asks candidates how they'd feel if stock went to zero — generativist · 2026-08-25
- Open source models lagging: Anthropic's internal models far ahead of delayed Astra — haider1 · 2026-08-25
- Eric Mitchell leaves Cursor after working on Composer — ericmitchellai · 2026-08-25
- AI Agents will consolidate SaaS around Systems of Record, not replace them — joshgonsalves_ · 2026-08-25
- Sora shutdown sparks thoughts on AI video economics: heavy users lose the most money — gandamu_ml · 2026-08-25
- Discussion: Do you still read your own code? — talkaboutdesign · 2026-08-25