
Inflection points: Opus 3 and 4.5

Why AI is going vertical (again) | Dianne Penn (Anthropic)
Topic 2 of 18
Published Jul 26, 2026 · Clip 8:57–13:51 (4:54)
Diane identifies training and testing Opus 3 as the moment a sub-200-person Anthropic committed to building a frontier model and answering the core question 'why should somebody choose Claude.' Teams across inference, research, fine-tuning, and pre-training rallied over a December winter break — working from childhood homes — building trust between product and research that still underpins their collaboration. She also recalls spotting in 2023 that people were using models for long-form code rather than just autocomplete; a relatively small training change ended up differentiating early Claude and attracting developer enthusiasts. A year later, Opus 4.5 landed during another winter break, this time with a product vehicle in Claude Code, letting users experience frontier intelligence running end to end in an agentic manner. Her synthesis: you need frontier products for frontier models, and the model and product enabled each other's inflection.